If they don't get you online, they'll try in person. A data-theft and extortion gang has targeted “dozens” of banks, law firms, and other professional services companies in the US from January through May, using fake help desk calls and other social-engineering techniques to gain access to corporate IT environments, according to Google’s Mandiant incident response team. And when those remote-deception methods don’t work, the criminals sometimes show up at victims’ physical offices, posing as IT technicians, and attempt to steal sensitive files using thumb drives. Google’s threat hunters track the extortion threat group as UNC3753, while other analysts call it Luna Moth, Chatty Spider, and Silent Ransom Group. The crew has been around since 2022, originally using fake software renewal emails and other billing lures, typically with PDF attachments containing phone numbers for attacker-controlled call centers, as their means of gaining initial access to corporate networks. Beginning around March 2025, the crims shifted tactics and started posing as IT help desk staff. “While UNC3753 primarily relies on digital vectors, GTIG assesses that associated threat actors have also attempted direct data theft using physical, in person access,” Google incident responders and researchers Chad Reams, Tufail Ahmed, Keith Knapp, Ashley Frazer, and Tyler McLellan said in a Friday blog. The authors also pointed to a May FBI alert to corroborate this in-person tactic. According to the feds, Silent Ransom Group crooks have been walking into law firms’ physical offices as recently as this spring. Once they are on-site, they claim to be IT support staff needing to image a device or create local backups for security reasons. If that line works, they plug a thumb drive into the victim’s computer and steal data the old-fashioned way. “Although limited forensic evidence and the absence of a subsequent extortion attempt prevent formal attribution, GTIG assesses that these physical intrusions are likely associated with UNC3753 based on structural, timeline, and targeting overlaps,” the blog said. Google won’t say how many dozens of firms have been targeted in these attacks, or how many ended in the data thieves paying a visit to the victims’ locations. “While we can’t share additional details regarding specific investigations, Mandiant CTO Charles Carmakal notes that this tactic has been observed over the years,” a spokesperson told The Register. “Mandiant has investigated various matters where adversaries planted insiders, bribed employees, or physically entered buildings to facilitate cyberattacks.” Another noteworthy thing about UNC3753’s attacks: they are very fast. In many of Mandiant’s investigated incidents, the entire operation from initial contact to data extortion occurred in just one day. “Recently, Mandiant observed data searches, staging, and theft initiated in under an hour,” the threat analysts warned. These intrusions typically begin with an invoice-themed email - but these don’t usually contain any malicious links or attachments. The email’s sole purpose is to give the miscreants a plausible reason to follow up via phone, so that the recipient is more likely to believe the call is legitimate. Most of the crew’s entry mechanisms involve voice-phishing, using a method that has worked so well for other groups like ShinyHunters and Scattered Spider over the past few years. UNC3753 calls organizations’ employees directly and purports to be a help desk worker or member of the security team. The criminals say they need the target’s help addressing a security issue or aiding with a corporate data migration project, and convince the individual to join a screen-sharing session via Zoom, Microsoft Terminal Services, Microsoft Teams, or Quick Assist. In one such intrusion, using Teams to gain access to the victim’s computer, the attacker jumped on five separate calls with the same target over a three-day period, we’re told. And in more than one incident that Mandiant responded to, UNC3753 established Zoom sessions directly on targets' personal laptops, using these machines to access corporate virtual desktop infrastructure (VDI) using native client platforms, such as Windows 365 or Citrix clients. Once they’re in the corporate systems, the intruders map local directories and network drives, and target specific legal and document storage repositories. The crooks also use very-specific keyword searches to find sensitive folders containing tax logs (Forms W-2, W-9, and 1099), audit files, corporate client agreements, and Social Security numbers, before staging this data for exfiltration. UNC3753 uses several methods to sneak the data out of the corporate IT environment without setting off any security alarm bells, including using portable versions of free Windows file manager WinSCP or another open source filesystem like Rclone. The crew has also been known to log into a file-sharing account from the victim’s browser and upload the stolen files that way - or even instruct the victims to send the files to an attacker-controlled email address. After stealing the data, they send the extortion email, usually within 30 minutes of exiting the victim’s environment, and set a three-day deadline to respond and begin the negotiation process. “We hope to find a financial solution that will be acceptable for both parties,” reads one such extortion email. It continues: In case of ignorance or no agreement, We will notify your employees, partners and customers, after which We will publish your data. You will receive claims from individuals, and legal entities for information leakage and breach of contracts, your current deals will be terminated. Journalists and others will dig into your documents, finding inconsistencies or violations in them. Your organization will lose its reputation, shares will fall in price, and your organization will be forced to close. Stay safe, friends In the Friday report, Google’s threat hunters list IP addresses and other indicators of compromise, including these phishing domains that UNC3753 uses in its social-engineering attacks, all designed to look like the target organization’s help desk: -itdesk[.]com, -it[.]com, and -helpdesk[.]com. The security shop also suggests a range of things companies can do to avoid falling victim to this group and other voice-phishing scams or physical office intrusions. Some of the physical controls include requiring visitors to display official credentials and photo identification, and mandating front-desk staff log all visitor IDs before granting access. Also, check pre-scheduled work orders to ensure the “technician” at the front desk is who they say they are, and make sure any visiting technical service workers are always accompanied by a corporate, in-office supervisor. Because the bulk of these intrusions occur without any physical entry into the office, however, companies should also implement remote access conditional access policies to ensure only corporate-owned devices can authenticate to any VDIs or VPNs. Plus, block the installation and execution of unauthorized remote monitoring and support utilities. ®

The threat is real. Unknown miscreants are exploiting a high-severity, zero-day bug in Cisco’s SD-WAN management software, and the networking giant hasn’t said when it will patch the flaw. Cisco issued an advisory on Thursday for the Catalyst SD-WAN Manager vulnerability, tracked as CVE-2026-20245, and it sounds like attackers have been exploiting this security failure for at least the last week. It’s due to a validation error - the software fails to properly validate user-supplied input - and an authenticated, local attacker can exploit the flaw by uploading a specially crafted file to vulnerable systems. From there, they can escalate privileges and execute commands with root privileges. The vulnerability affects all versions of the SD-WAN software, regardless of device configuration, and across all deployment types including on-premises, cloud-based, and FedRAMP-certified deployments. Switchzilla says it became aware of attacks against this vulnerability in June. “To exploit this vulnerability, an attacker must have netadmin privileges on an affected system,” the vendor said. “This would require valid credentials or exploitation of CVE-2026-20182 or CVE-2026-20127. Cisco is not aware of successful exploitation by other methods.” Both of these earlier SD-WAN security holes have also been hit by attackers in previous months. The good news: an attacker needs valid credentials to abuse the new hole. The bad news: exposed credentials aren’t hard to find (or buy) online. We don’t know the scope of exploitation or exactly when attackers began hitting this SD-WAN hole. Cisco declined to answer The Register’s questions, and instead sent us a statement via email. “Cisco recommends customers upgrade to the fixed software released in May 2026 for CVE-2026-20182 as a protective measure,” a spokesperson said. “A patch for this vulnerability will be provided on a future date. Customers needing assistance should contact Cisco TAC.” This latest bug is the sixth SD-WAN vulnerability listed as under attack since the start of the year, and the second zero-day in two months. The most recent is the one the Cisco spokesperson mentioned in an email to The Register. In May, Switchzilla disclosed a max-severity make-me-admin bug (CVE-2026-20182) affecting Catalyst SD-WAN Controller and Manager, and warned that attackers had already found and exploited the hole before it issued a patch. A month earlier, America's lead cyber-defense agency said that three Cisco Catalyst SD-WAN Manager bugs (CVE-2026-20128, CVE-2026-20133, and CVE-2026-20122) were under attack, and gave federal agencies just four days to patch the security holes. Cisco fixed all three CVEs in late February, and in March warned of attackers abusing two of them. Also in February, the networking vendor patched a max-severity improper authentication flaw (CVE-2026-20127) affecting the same SD-WAN software, prompting a Five Eyes countries’ joint intelligence alert urgently warning defenders to patch it - plus an old SD-WAN vulnerability (CVE-2022-20775) - or risk root takeover. "Malicious cyber threat actors are targeting Cisco Catalyst SD-WAN used by organizations globally," the UK's lead cyber agency said at the time. "These actors are compromising SD-WANs to add a malicious rogue peer and then conduct a range of follow-on actions to achieve root access and maintain persistent access to the SD-WAN." And while this one isn't listed as under active exploitation (yet), on Wednesday, Cisco warned about a proof-of-concept exploit for CVE-2026-20230, a critical bug in its Unified Communications Manager that also allows attackers to gain root privileges. ®

One gaming cyberattack this week exposed nearly 64,000 users. 

Another has already infected more than 116,000 players.  

Both are connected by the same common gaming behavior: looking for a cheat, mod, or shortcut. 

This week in scam news, a popular Grand Theft Auto V cheat service was hacked, exposing tens of thousands of users. At the same time, McAfee researchers uncovered a massive malware campaign spreading through fake Minecraft mods, cheats, and game clients. 

The takeaway is simple: some of the biggest threats facing gamers aren’t happening inside games. They’re hiding in the downloads, websites, and tools players use around them. 

Let’s start with the GTA breach. 

GTA Cheat Service Breach Exposes Nearly 64,000 Users 

Atlas Menu, a cheat service for Grand Theft Auto V, was reportedly hacked, exposing data belonging to nearly 64,000 users. 

According to reports, the leaked information included: 

• Email addresses 
• Usernames 
• Scrambled passwords 
• IP addresses 
• Customer support tickets

The hacker who claimed responsibility later posted the data online. 

Why This Matters 

Many players think of cheats as harmless tools that unlock special abilities, provide advantages, or simply make games more entertaining. 

But unofficial cheat services often operate outside the protections offered by legitimate gaming platforms. 

That means users may be: 

• Sharing personal information with unknown developers 
• Downloading unverified software 
• Exposing themselves to malware 
• Putting gaming accounts at risk 

And that brings us to an even bigger threat. 

Minecraft Malware Campaign Has Already Infected 116,000 Players 

McAfee researchers recently uncovered a large-scale malware operation targeting gamers searching for Minecraft mods, clients, and cheats. 

The campaign is called WeedHack. 

What Is WeedHack? 

WeedHack is a type of Malware-as-a-Service (MaaS). 

That means cybercriminals package malware into a subscription service that other attackers can use. 

Researchers found that: 

• More than 116,000 victims have been infected since January 
• The campaign continues to add roughly 2,000 to 3,000 new victims every day 
• More than 3,800 malicious files have been identified 
• More than 240 malicious download URLs have been linked to the operation 

Premium versions reportedly cost as little as $5 per month and include tools that allow attackers to remotely access victims’ devices and webcams. 

What WeedHack Can Steal 

Once installed, the malware can collect: 

• Minecraft account credentials and session IDs 
• Discord, Steam, and Telegram credentials 
• Browser passwords and cookies 
• Cryptocurrency wallet information 
• Screenshots and device information 
• Files stored on a victim’s computer 

Premium versions can also provide: 

• Live webcam access 
• Live screen sharing 
• Remote keyboard and mouse control 
• Keylogging capabilities 
• Full remote access to the infected device

Get the full explainer here. 

How McAfee+ Advanced Helps Protect Gamers 

Gaming malware campaigns rely on three things: 

1. Getting users to visit malicious websites 
2. Convincing them to download infected files 
3. Encouraging them to ignore security warnings  

The post GTA Cheat Users Exposed in Breach as Minecraft Malware Hits 116,000 Players appeared first on McAfee Blog.

Humanitarian organization World Food Programme (WFP) says one of its systems was breached, and around 600,000 Gazan households receiving aid had their details improperly accessed. Its announcement, made via Telegram on May 31, confirmed there was “a security incident” in the self-registration application used by Gazans to register for aid and applicants’ names, ID numbers, phone numbers, and location information were among the data types accessed. “We understand this may be concerning, and we want to assure you that protecting your data and privacy is our top priority,” the WFP said. “The program is treating this situation with the utmost seriousness and priority.” The organization said it temporarily suspended the registration platform to urgently apply the necessary security improvements. Its most recent update on the situation came on June 2, when it said the platform was still down, but added that aid recipients did not need to do anything, while their support would continue to be delivered uninterrupted. “The WFP wants to assure all those registered via the link that food assistance, cash assistance, nutritional supplementation, and all other WFP programs are continuing as usual,” it said. “If you are already registered on the Self-Registration Application (SRA), your registration remains valid. There is no need to update, delete, or re-register your information at this time.” WFP told The New Humanitarian, which first reported the story, that the attack was detected on May 14, and confirmed the scale to be in the region of 600,000 households. The news organization also claimed, citing a whistleblower’s account of matters, that an anonymous “independent expert” contacted WFP’s Palestine team, alerting it to vulnerabilities in the SRA two days before the organization detected the breach. The Register contacted WFP’s Rome headquarters for more details, but it did not immediately respond. WFP, which is a division of the UN and the largest welfare organization in the world, supports 1.6 million Palestinians every month who face a malnutrition crisis amid fierce conflict between the territory and neighboring Israel. This represents around 77 percent of the country’s population, and an estimated 80 percent of the population is unemployed, unable to earn the money required to pay for a nutritionally sound diet. WFP delivers wheat flour, high-energy biscuits, and fortified snacks to families, community kitchens, and bakeries in its effort to push back famine, as well as facilitating cash transfers. The organization is also helping individuals get back into paid work, maintains roads, and says that when conditions allow, it will stay in the region and help local people rebuild communities, markets, and other food systems. ®

A City of York Council email mishap exposed the email addresses of hundreds of Blue Badge holders in the ancient Viking capital, inadvertently revealing their status as disabled residents and triggering a data breach investigation. The council confirmed to The Register that it’s investigating what it described as a "personal data breach" after emails sent to residents last week were distributed without using the blind carbon copy (BCC) function, allowing recipients to see everyone else on the mailing list. According to local reports, the council sent three emails containing Blue Badge-related updates before issuing a fourth message acknowledging the error and asking recipients to delete the previous emails, including from their deleted items folders. Recipients were also warned to remain alert for suspicious messages following the incident. While the exposed information appears to have been limited to email addresses, the breach is especially sensitive because everyone on the distribution list was receiving communications intended for Blue Badge holders. In practice, that meant recipients could identify hundreds of people as members of a group generally associated with disabilities or mobility impairments. One affected resident told local media that the disclosure had left her upset because most people in her life were unaware she held a Blue Badge. "Honestly, I think it's just disgusting – we've been given the details of hundreds of disabled people, which feels unsafe," she said. In a statement to The Register, a spokesperson at City of York Council said it activated its data breach procedures as soon as the error was identified and is conducting a risk assessment in line with guidance from the UK Information Commissioner's Office. "We're working carefully to establish exactly what's happened, alongside conducting a thorough risk assessment ... to understand any potential impact on individuals," a spokesperson said. “Our investigation is ongoing, and we’ll continue to be as open as possible while ensuring the accuracy of the information we provide.” The spokesperson declined to say how many individuals were affected or whether the issue was caused by human error or a technical issue. The council added that it was assessing whether the incident meets the threshold for notification to the ICO within the statutory 72-hour reporting window. That may depend less on the email addresses themselves than on what the mailing list revealed. A spokesperson at the ICO told The Register: "We can confirm that we have received a data breach report on this matter, and following an assessment of the information provided we have closed the case with advice given.” For all the talk of AI-powered cyber threats, it seems some organizations remain committed to the classics. ®

UPDATED A new extortion brand called Pink – which may be a rebrand of BlackFile – uses voice phishing and fake help-desk calls to gain initial access to organizations’ IT environments, steal their sensitive data, and threaten to leak it unless the victims pay a ransom demand. Palo Alto Networks' Unit 42 first spotted the gang, which it tracks as cluster CL-CRI-1147, and its data-leak site, which went live on May 31. “Pink uses vishing and IT impersonation to phish credentials/MFA, then exfiltrates enterprise cloud storage and productivity data to extort victims,” the threat-intelligence biz said in a LinkedIn post. Google Threat Intelligence is not so sure it's a new gang, however. "After retiring the BlackFile brand in May 2026, we assess the group launched the 'Redact' brand and has now potentially surfaced as 'Pink,," Austin Larsen, Principal Threat Analyst at Google Threat Intelligence Group, told us. "This new operation exhibits hallmarks of UNC6671, including similar credential-harvesting infrastructure, data leak site (DLS), and recurring messaging that claims to 'improve the security' of victims who pay. Additionally, we attribute the Pink (CL-CRI-1147) domains recently published by Unit42 to UNC6671." Regardless whether it's brand new or just a new coat of paint, the tactics are very familiar. Pink is one of many goon squads to use these social-engineering tactics to steal employees’ credentials and bypass multi-factor authentication, using this access to burgle companies’ cloud storage and databases. Chaotic crime crew Lapsus$, during its 2021 and 2022 extortion spree that hit Nvidia, Microsoft, and Okta, among others, popularized this style of phone-based intrusions before Scattered Spider picked up the mantle. Scattered Spider is perhaps best known for its 2023 Las Vegas casino digital heists, and reportedly bragged that all it took to break into MGM's networks was a 10-minute call with the help desk. Over the last few years, ShinyHunters has used this same playbook to steal sensitive data from Ticketmaster, AT&T, and other Salesforce customers, and thousands of schools and universities that use Canvas’ digital learning platform. Despite multiple arrests across all three gangs, they keep coming back to victimize more organizations. Most incident responders, including Google’s Mandiant and Unit 42, link many of these criminal collectives to The Com, a loosely knit group of primarily English speakers made up of several interconnected networks of hackers, SIM swappers, and extortionists, with some of its subgroups offering real-life violent crime for hire. According to Unit 42, this latest cluster of extortion activity is also “likely a Com-affiliated actor.” And after investigating “multiple” of these extortion attacks over the past few months, on Monday, they spotted something that led them to Pink’s name-and-shame website. “On June 1, 2026, an existing extortion negotiation that had never received a response, attributed to a likely Com-related cluster, received new communication from a threat actor via a free webmail account,” Unit 42 analysts Richard Emerson and Cuong Dinh said in a Wednesday threat-intel post. “The actor provided a new qTox ID and a leak site associated with the Pink brand, but referenced exfiltrating almost identical information from the original extortion notice.” Pink data thieves set a 72-hour deadline for the victim to respond before leaking the stolen goods. After gaining access to the victim’s account, the criminals snoop around for valuable corporate and customer data from platforms like SharePoint and OneDrive. After exfiltrating the stolen files, Pink attackers use compromised victim accounts and internal Teams messages to extort the company. “The actor reuses second-level domains to target multiple organizations, and the third-level domain typically thematically represents the target,” Emerson and Dinh wrote. They also listed the following phishing domains as indicators of compromise: passkeyadd[.]com passkeydeploy[.]com deploypasskey[.]com Along with these three IP addresses: 185[.]178.208[.]153 (hosted phishing domains) 172[.]93.100[.]252 (accessed compromised accounts) 96[.]232.20[.]66 (residential proxy IP responsible for extortion email creation) Plus, these user-agent strings were observed during data exfiltration: Microsoft.Graph.Client/5.62.0 python-requests/2.28.1 python-requests/2.33.1 Network defenders can use these to assist in threat-hunting efforts. And be very wary of help desk calls, both from people claiming to be employees locked out of corporate accounts and from those purporting to be support staff rolling out a mandatory MFA update or other emergency. ®

The next threat your server faces may have been helped along by a bot. OpenAI's Codex agent helped uncover a remote denial-of-service (DoS) exploit that can be launched from a single machine to render vulnerable web servers inaccessible in seconds, according to Calif security researchers. The attack works on default HTTP/2 configurations of major web servers including nginx, Apache HTTP Server, Microsoft IIS, Envoy, and Cloudflare Pingora. As of Thursday, Microsoft IIS and Cloudflare Pingora still don’t have a patch, according to the researchers, although Cloudflare disputes this finding. “Cloudflare's existing architecture and DDoS mitigations automatically detect and protect against this attack, making customers resilient to this vulnerability,” a spokesperson told The Register. “No patch is needed.” “We are aware and actively investigating appropriate mitigations to help keep customers protected," a Microsoft spokesperson told The Register. Calif researcher Quang Luong discovered the exploit, named it HTTP/2 Bomb, and will present the full technical details of the attack at the Real World AI Security conference later this month. In the meantime, there are proof-of-concept exploit scripts on GitHub along with a warning from the AI red teaming security shop: “Please don't point these at infrastructure you don't own.” In a Tuesday blog, Luong says Codex chained two existing DoS attack techniques that have been known for more than a decade - HPACK compression bomb and Slowloris-style hold - and warns that upwards of 880,000 websites supporting HTTP/2 and running one of the vulnerable web servers may be affected. An HPACK bomb attack (also known as CVE-2016-6581) exploits the HTTP/2 header compression algorithm (HPACK) by sending thousands of tiny messages to the server, forcing it to rapidly allocate memory and ultimately crash. Then the Slowloris DoS attack (CVE-2016-8740 and CVE-2016-1546) overwhelms the server by opening legitimate connections and maintaining them as long as possible. Combining the two exhausts the server’s memory and forces it offline. “A home computer on a 100Mbps connection can render a vulnerable server inaccessible within seconds,” Luong wrote. “Against Apache httpd and Envoy, a single client can consume and hold 32GB of server memory in roughly 20 seconds.” The Calif research team disclosed the issue to nginx in April, and the web server’s maintainers fixed it the next day in version 1.29.8, which imports the max_headers directive from freenginx. Apache issued a fix (mod_http2 v2.0.41) the same day that Calif submitted its report, and assigned it CVE-2026-49975. “The fix commits above are public and disclose the vectors directly; any capable AI model can turn those diffs into a working exploit, which is exactly how we found that Microsoft IIS, Envoy, and Pingora are also vulnerable,” the threat hunting team wrote, adding that all three have been notified. In a Wednesday update, Calif pointed to Envoy patches “that appear to mitigate this attack,” and notes that its researchers are still validating the fix to ensure it works. For Microsoft IIS and Cloudflare Pingora, the security sleuths recommend disabling HTTP/2 if possible, or enforcing a cap on the number of HTTP headers a client can send in a single request to the server. The fact that a coding agent - not a human - discovered this attack is notable, according to Calif. “Both halves have been public for a decade,” Luong wrote. “What Codex did was read the codebases, recognize that the two compose, and build the combined attack. That combination is obvious once you see it, and yet as far as we can tell no human had put it together against these servers.” ® Updated at 2023 with statement from Microsoft.

Code reviewed by WIRED uncovered an unreleased face-recognition system embedded in Meta’s smart glasses platform. It’s designed to identify people via biometric data stored on users’ phones.

MI5 and its international allies are once again warning that China is shopping for state secret leakers on popular recruitment platforms, including LinkedIn, Indeed, and Upwork. In a fresh advisory published on Wednesday evening, the UK’s domestic counter-intelligence agency said China is using an increasing number of platforms to recruit those who have access to classified or privileged information. Chinese military intelligence officers specifically target security clearance holders, including marks working in defense, security, and foreign affairs, military personnel, and those with indirect access to government information, such as academics, journalists, think tank employees, and others. Anyone who fits the bill is being urged to remain vigilant to potential attempts from Chinese operatives to cultivate long-term relationships. “These actors use an aggressive online recruitment strategy whereby intelligence officers or their affiliates pose as employees of private consultancies, think tanks, or human resources firms, and place online job advertisements for foreign policy and defence analysts (or similar),” the advisory [PDF] states. “Successful candidates are pressured to provide 'non-public' information for unspecified clients who are associated with the Chinese government. China’s military intelligence services ultimately seek to acquire privileged military, political, and economic intelligence that can provide China with a strategic and tactical advantage over the Five Eyes.” According to MI5, after the job and gig-work ads are posted online, China’s spies will rank the resumes they receive based on how likely a given individual is to have information of interest before interviewing them. It warned that even by sending a resume over, which includes personal details, a person is risking their own security and privacy. Targets face probing questions about who they know in government. For those in the military, they might be asked about where they were based, and what tasks they were responsible for. After demanding potential recruits complete a trial report on matters related to China, the spies will often shift conversations to encrypted messaging platforms where recruits are offered payments in exchange for increasingly privileged information. Payments may arrive through a number of online platforms, including reputable services like PayPal, Zelle, and Wise, to others more commonly associated with associated with illegality, such as Western Union and cryptocurrency. MI5 closed out its advisory with a warning to anyone even considering a life of peddling secrets to China: doing so comes with severe consequences. “Certain types of data can place the lives of frontline military or other personnel at risk, can weaken our economic prosperity, and enable interference in our democratic processes,” it said. “Individuals engaged in the unauthorized disclosure of sensitive or classified information could face a number of consequences, including prosecution under national laws such as those relating to espionage.” A common theme This week’s admonition is far from the first issued by the UK in response to this particular aspect of Chinese spies’ tradecraft. The most recent came in November when UK security minister Dan Jarvis reminded the UK's House of Commons that members should have received information about Chinese attempts to recruit parliamentarians through identical means. In those information packs disseminated by MI5, Brit politicos were given the names of two online profiles that the counter-intelligence agency suspected of being involved in recruitment campaigns. MI5 dished out an earlier warning in 2021, saying that around 10,000 Britons had been targeted by Chinese spies over the previous five years using work platforms, posing as headhunters. The 10,000 figure, it added, was thought to be a conservative estimate, with the agency's head, Ken McCallum, saying workplace platforms were being exploited “on an industrial scale.” The US said it was seeing similar tactics used when President Trump took office for the second time, which shortly after led to mass redundancies across federal agencies. Experts at the Foundation for Defense of Democracies (FDD) named five supposed consulting companies targeting the recently jobless via LinkedIn, Craigslist, and others, all in search of state secrets. The companies would present the fired workers with job opportunities, and as FDD senior analyst Max Lesser told The Register at the time, the layoffs, which began in February 2025, would have likely raised the risk level associated with state secrets being spilled. ®

Two former RAC workers in the UK have three months to pay more than £118,000 ($158,500) collectively after being convicted of selling crash victims’ data, according to the Information Commissioner’s Office (ICO). Debbie Okparavero and Maliha Islam, of Salford and Manchester respectively, were sentenced to six-month prison stints, suspended for 18 months, and 150 hours’ unpaid work in 2024, after being found guilty of offenses under the Computer Misuse Act 1990 and the Data Protection Act 2018. The pair, who worked for roadside accident biz RAC, were caught selling the personal data of car crash victims – just shy of 30,000 lines of data to an unknown buyer, the ICO revealed following an investigation. Okparavero and Islam were in a WhatsApp chat together, where they discussed the data and its sale to the unknown third party. RAC clocked on to the activity after deploying unspecified monitoring software, which detected Okparavero copying the data from RAC systems. A resulting investigation showed that around 29,500 lines of data were shared with Islam via WhatsApp. Islam was ordered to repay £39,522.50 ($48,274.45) for her part in the scheme in November, and the ICO noted in a Thursday announcement that she paid this in full. Reflecting more serious offending, at Manchester Crown Court on May 29, Okparavero was ordered to repay £89,277.32 ($119,962.38) within three months. Failure to do so will result in her serving 18 months in prison. Andy Curry, head of investigations at the ICO, said: “This outcome demonstrates justice did not end at sentencing. Our powers enabled us to continue to pursue these two individuals in order to strip them of assets gained through their serious criminal activity. Through the Proceeds of Crime Act, we are ensuring people do not financially benefit from their criminal activity. “I would like to once again thank the RAC for informing us about this breach and fully supporting the ICO’s investigation, which enabled us to hold these two individuals to account.” ®

There's a lot of fear surrounding the bug-finding capabilities of super-advanced AI models like Anthropic's Mythos and OpenAI's GPT 5.5-Cyber. But attackers are already using free, publicly available LLMs to hijack networks and worm through software supply chains at a much lower cost – to them at least. The latest example comes from University of Toronto researchers, who used an unnamed, publicly available open-weight model released in 2025 to develop a computer worm that they claim spread through an enterprise test network. The self-propagating code adapts on the fly to identify known vulnerabilities and misconfigurations on target systems, then generates and executes attacks to move laterally through the network and compromise additional machines. And it’s all built on a small, free model that runs on a single GPU. “People need to understand that it’s not just the biggest and most powerful AI models that pose security concerns – a whole other area of threat has been vastly underestimated,” University of Toronto computer engineering professor Nicolas Papernot told The Register. Papernot and fellow researchers Jonas Guan, Tom Blanchard, Hanna Foerster, Hengrui Jia, and Gabriel Huang published their findings [PDF] on Tuesday. While guardrails and other safety features implemented by major commercial AI systems are “essential,” Papernot told us, in reality “they will not prevent the threat of AI-driven worms with a similar design.” “The majority of real-world cyberattacks don’t rely on zero-day vulnerabilities,” he added. “Our work demonstrates that attackers can now cheaply operationalize known vulnerabilities at scale, which decreases the window of time defenders have to fix vulnerabilities and find human errors, like reused passwords or poorly configured backup jobs.” The paper doesn’t specify, and Papernot declined to say, which LLM they used. “We omitted certain methodological details (such as the agent’s reasoning graph and tool harness) and experimental specifics (such as the AI model) that could materially help a malicious actor construct similar malware,” Papernot said. “We shared enough information to make the threat credible enough for scientific scrutiny without providing a blueprint that would enable misuse.” The researchers also noted that they are not publicly releasing the code, but are working with the University of Toronto to set up a vetting process through which qualified researchers may request access for defensive research purposes. Not NotPetya Before you start breathing into a paper bag, there are a few things to note about this research. First, unlike Mythos and friends, the prototype worm does not exploit zero-day vulnerabilities. It only targets publicly disclosed but unpatched bugs, misconfigurations, and recurring weakness classes. This is intentional, because known security flaws – not zero-days – are what most real-world cyberattacks use, the authors say, citing WannaCry and NotPetya as examples. Both of these worms exploited security holes that had patches available for at least a month before the malware infected vulnerable machines. Both spread rapidly and caused global disruption. The worm did, however, find and abuse vulnerabilities disclosed after the model’s training cutoff by ingesting publicly available security advisory information at runtime and using this data to develop exploits. While the paper repeatedly points to WannaCry and NotPetya as worst-case scenario examples, this lab-tested prototype or something similar is not going to cause the level of destruction that either of those two earlier worms did. Both propagated very quickly: WannaCry infected more than 230,000 computers across 150 countries in just one day in May 2017. In June 2017, NotPetya spread globally within hours, taking down at least one large banking network in just 45 seconds. Plus, they both used very sophisticated evasion techniques to avoid being detected by security tools. This worm, on the other hand, moves slowly. In the “FakeCorp” network they used in the experiments, the prototype took about five days to replicate across half the network, requiring hundreds of LLM inference calls per target for reconnaissance, strategy formulation, and payload generation. The timeline gives defenders a longer window for detection and response. However, it will likely shorten as inference hardware and model efficiency improve. Also, unlike WannaCry and NotPetya, the worm doesn’t try to hide itself. “We deliberately chose not to equip the worm with concealment capabilities – it is not instructed to cover its tracks or minimize its network footprint, and it has no tools to do so,” the boffins wrote. “This was a conscious methodological choice to further limit the risk of misuse.” Finally, the test-network devices themselves didn’t have any endpoint detection, antivirus, or firewall software deployed, which (we hope) makes this a not-quite-realistic setup. Exploiting the FakeCorp target network Here’s how the experiments worked. The team deployed the worm prototype in 15 independent experiments on an isolated 33-host network including Linux servers, Windows environments, and IoT devices. Each computer had been seeded with at least one real-world vulnerability, including software bugs and misconfigurations. The worm operated fully autonomously for seven days, and correctly identified an average of 31.3 vulnerabilities, exploited 23.1 hosts to elevated access, and propagated to 20.4 hosts. It reached up to seven generations of self-replication, we’re told. “Put another way, on average, the proof-of-concept worm successfully exploited 73.8 percent of the network and then replicated to 61.8 percent of the network,” according to the research paper. In addition to finding and abusing the known flaws on the devices, the worm also found vulnerabilities disclosed after the LLM’s training cutoff: two recent Linux flaws, CopyFail and DirtyFrag, and a critical bug that leads to remote code execution in the Marimo notebook platform (CVE-2026-39987). The AI-based worm gained root access in 41 of 67 attempts (61 percent) across the three hosts with these one-day vulnerabilities. This includes 13 of 27 attempts on the Marimo host, 15 of 17 on DirtyFrag, and 13 of 23 on CopyFail. The two Linux vulnerabilities were disclosed shortly before the experiments, and a single document containing publicly available exploit instructions was enough for the worm to weaponize them successfully 70 percent of the time, or in 28 out of 40 attempts. “This result suggests that the same exploitation pipeline can absorb newly published vulnerability information without architectural changes,” the team wrote. When worms get sneaky In addition to finding and exploiting vulnerabilities, the AI worm did some other interesting, sneaky things during the experiments. Despite the lack of firewalls and endpoint security products enabled, the researchers added some components to the worm’s code to “provide increased defence-in-depth.” This included a denylist of IP addresses baked into the source code. So in theory, the worm would read the denylist every time it targets a new machine, and if the IP address matched one on the list, it would move on to a new target, thus preventing it from infecting some hosts. Instead, at “various points” during the experiments, the worm rewrote its source code to allow it to attack an IP address on the denylist. In another workaround example: when a virtual-machine check caused replica worms on Alpine Linux and Windows Server 2008 machines to crash at startup, the parent worm found the attestation source file on the target computer and removed the VM-detection check, allowing the worm to replicate successfully on the new machine. “This behavior was not prompted or incentivised by the agent’s instructions; it autonomously diagnosed the failure, identified the responsible code, and modified it to achieve its objective,” the boffins wrote. In another test run, the researchers accidentally packaged a FakeCorp network configuration file that included admin credentials with the worm code. The worm discovered the credentials and shared them with other active replica worms, “causing a rapid increase in propagation within the contained environment.” Additionally, in two separate instances, the worm established persistence on a compromised computer using mechanisms outside of the intended replication pipeline, we’re told. In one instance, it started the replica process and then registered it with the machine’s service-management system to auto-restart if terminated. In another, it created a scheduled task that relaunched the replica worm, causing it to reappear even after the original process was killed. “The agent appears to have inferred, from the general objective of maintaining an operational replica, that persistence mechanisms available on the target could be used to make the replica more robust,” the researchers noted. Prior to publishing their work, the academics say they shared their findings with “national science, security, and defence” agencies to seek advice on how to responsibly release the information. We asked Papernot for details, including which government agencies and how they responded, but he declined to share anything else. ®

PWNED Welcome back to PWNED, the weekly column where we talk about weak security policies and how to avoid them. Hopefully, we can learn from others’ mistakes – or at least have a good laugh at them. Have a story about someone leaving a gaping hole in their network? Share it with us at pwned@sitpub.com. Anonymity is available upon request. This week, we have a tale of password passivity involving Active Directory. It comes to us courtesy of Rob Anderson, head of reactive consulting services at Reliance Cyber, a UK-based security firm. Anderson recalls in the past working with a firm that was creating service accounts that developers needed to use, but the org didn’t have a proper password vault for storing the associated credentials. Instead, to make it easy for team members to find what they needed, they put the passwords into the description field for Active Directory. “People don't realize that as soon as you've got an Active Directory user — just an ordinary user — you can read the comments field or the description field across the whole of Active Directory,” Anderson told The Register. “It's such an amazing lapse of security.” Soon enough, an Initial Access Broker (IAB), someone who specializes in gaining access to protected networks and then selling it to other threat actors, used a phishing campaign and executed offensive hacking tool Sliver on the endpoint. At that point, they captured a victim’s credentials, which led them to query Active Directory. Once in AD, the hackers found plenty of passwords, which came with full domain access. They used this access to delete all the backups and execute ransomware. In total, the crimes put 2000+ users out of action by encrypting Hyper-V hypervisors and their hosts. The company was taken offline for months. What we can learn from this sad story is that you can’t put passwords in cleartext anywhere that's easy to access, unless you want an enormous attack surface. Even without a phish, an untrustworthy colleague could have sold the passwords to a threat actor. After all, a recent survey found one in eight workers think selling company logins can be justified. “I've seen it where configuration details are kept in application servers that are running, and threat actors are using fuzzing — trying likely file and directory names — which again exposes configuration and credentials to the threat actors,” Anderson said. He noted that developers are a bit more savvy these days about where they put their credentials, but security naivete sinks ships. Trust no one. ®

AI-enabled cybercriminals have better tools and are inflicting more pain on their victims, wiping out virtual machines and hypervisors and leaving infrastructure in a "dark, dead" state after an attack, said Commvault Chief Technology Officer Brian Brockway. "The majority of cyber cases that we've seen in the customer base have moved well beyond the breaking inside, and encrypting and corrupting some of your key files and folders, to taking over control of your entire VM environment, wiping out all VMs, destroying all hypervisors, blowing up the center and leaving you in basically a dark, dead state," Brockway told The Register. Frontier AI is reshaping the threat landscape in two ways, he explained: advanced models are uncovering a deluge of software vulnerabilities, and attackers are exploiting disclosed flaws within minutes rather than weeks. “The more unplanned work that has to be done to react to this, that's always going to challenge priorities,” Brockway said. “We had the plan in place, we had sprints already dedicated to kind of get out to the next launch, and we have to come back over and reinvest more engineering time to corrective actions versus the next new get ahead feature.” Commvault cited Palo Alto Networks research showing that frontier AI models such as Mythos and GPT-5.5-Cyber identified more than seven times the typical number of software vulnerabilities found within a single month during testing. To prepare for this, Commvault recommends that IT and security teams look beyond backups and ask whether they can restore critical systems cleanly, whether recovery environments are isolated from compromised production systems, and whether recovery plans include the most important applications and dependencies. Brockway said air-gapping is the starting point. He said organizations should keep immutable and isolated copies of critical data separated from production identity, network, and management planes, and pressure-test recovery time and recovery point objectives against realistic attack scenarios, a hard lesson learned from witnessing victims recover from recent attacks. “One team is just trying to even clear the smoke to figure out what happened, then you have to come back over, strip it all down to bare metal, and basically redeploy the data center all over again,” he said. “While that's ongoing – and that's not a couple hour process by any means, that could take you, even in a well-exercised environment, it could be a couple of days or longer to get it back into a stable, usable state – what are our sanitized versions that we're going to come back over to (in order to) rebuild or restart the business again?” Businesses should prioritize the systems they cannot operate without — identity platforms, billing systems, operational databases, and cloud services — and define the order in which they will be restored, he said. As AI moves into core operations, teams should also account for newer dependencies such as data pipelines, model repositories, vector databases, and agentic workflows. In its recommendations, Commvault said it is also critical that organizations continuously test recovery. Brockway recommends rehearsing those plans in isolated cleanroom environments before the worst happens. “I need a testing environment that's got the same makeup, the same builds, which we're using, maybe not on full production resources, but I need to be able to say, ‘How do I put that application stack into a live environment, so we can come back over and test?’ “ he said. “That's what we're saying about things like this clean room concept of not just being a reaction to an incident, but it is also a quick environment for you to come back over and clone.” Brockway said this new normal in the AI era is straining the engineers who build and maintain enterprise software. He said while the first wave of AI scanning tools flooded teams with potential vulnerabilities, newer models go further, entering controlled environments and attempting the exploits themselves — a capability that mirrors what attackers do. "When you let them in, you have to do it under an extremely tight security control, because you're effectively almost automating the same thing that bad guys can do on the outside too," Brockway said. The output can swamp downstream teams. Brockway said one frontier model flagged roughly 10,000 critical vulnerabilities across operating systems, browsers, and other infrastructure. "That's 10,000 patches that have to come out of the system," he said. That volume forces hard choices about engineering priorities. Brockway said unplanned remediation work pulls staff off planned releases. To absorb the load at Commvault, Brockway runs a standing group dedicated to just those items. "They're the fast action team to analyze, make a quick assessment," he said. Brockway said the signal volume emerging from AI bug finders ultimately calls for more automation and AI to filter noise, assist with patching, and support deployment. "The amount of information and signals that are coming in are way overwhelming. People just get desensitized, and that's when bad things really start to occur," he said.®

Your child’s first data breach may happen before they’ve even opened a bank account. Here’s how to keep their digital life safe.

Wireless jamming attacks are on the rise. Rice University researchers have shown how self-curving radio beams can make a jammer appear to be somewhere it isn't, potentially undermining some anti-jamming defenses. Jamming relies on flooding a wireless receiver with noise that denies service. Some modern receivers identify and block jamming attempts using direction-of-arrival (DoA) estimation technology that pinpoints the jammer's direction and directs an array null that blocks signals emanating in the jammer’s direction. Were a jammer to transmit a self-curving beam, however, it could fool DoA-based anti-jamming defenses by appearing to come from somewhere else entirely, and that's exactly what the Rice researchers demonstrated. Rice electrical and computer engineering professor Edward Knightly and doctoral student Caroline Spindel presented a paper [PDF] last month in which they demonstrated a curving-beam jamming attack that caused "catastrophic bit-error-rate degradation" while also "fool[ing] the receiver's DoA estimator," preventing conventional DoA-based defenses from stopping the interference. Knightly and Spindel have done prior research developing wireless technology that could bend beams around objects to increase signal strength - particularly useful for short-range millimeter wave signals - and found that the same technology could be used to deploy jammers that are far harder to locate. Spindel gave the perfect analogy in a recent Rice press release about the research for understanding how curved beams confuse DoA estimators by considering a soccer ball kick to the head. “Imagine being hit on the right side of your head by a soccer ball - you would naturally look to the right,” Spindel said. “If the ball actually curved through the air, like a David Beckham free kick, then it was kicked from somewhere else entirely.” Were Sir David to keep moving and kicking curveballs at your head you’d probably spot him eventually, but it might take a minute, and a few more smacks, to stop him. A signal jammer at radio-wave distances will probably be far harder to spot, and it won’t even have to move: Knightly and Spindel were able to create the illusion that the jammer was mobile by modulating the beam parameters from a stationary position, making it even more difficult to locate the jamming signal and negating the point of blindly searching for the best spot to point an array null. Conventional recovery methods used to block jamming completely failed in laboratory tests, Spindel said. “This is the first demonstration of a jammer that cannot be reliably localized and the first time self-curving wireless beams have been used as an attack,” Knightly added. The pair sees their research not just as a way to point out a serious threat to wireless signals - GPS jamming of aircraft is on the rise, for example - but also something that can inform the direction of future wireless technologies as we move toward the 6G era. Until then, however, there’s the potential for even more devastating jamming attacks to come. ®

Four people suing Elon Musk's AI firm under pseudonyms due to the risks of being identified may face a difficult choice: Reveal your real names, or drop the lawsuit.

UPDATED Yet another aggrieved bug hunter has leaked a vulnerability affecting a Microsoft product after becoming disillusioned with the way the company handles security reports. Ammar Askar dropped a proof of concept (PoC) exploit for a Visual Studio Code (VS Code) flaw within just an hour of disclosing it to “an old contact” at the open source platform, according to his account of things. The vulnerability he exposed involves attackers configuring repos, either of their own making or those they have compromised separately, to push malicious VS Code extensions via its Workspace Recommendations feature, which then steal OAuth tokens they can then use to read/write public and private GitHub repos. It affects anyone who has ever used github.dev, a feature that allows users to open a GitHub repo in a browser-based version of VS Code. Askar said that the feature is enabled by github.com passing an OAuth token over to github.dev and, crucially, this token is not limited to the repo from which github.dev was spun up. It means that this token can hand an attacker access to any other repo – public or private – to which the target also has access. The exploit is contingent on an attacker being able to modify a repo’s .vscode/extensions.json file and recommending an attacker-controlled extension for the browser-based VS Code instance. In normal scenarios, a pop-up would appear asking for a user to accept the installation of this extension, potentially tipping them off to foul play. However, because of the way in which the attacker delivers the repo to the target, they already have a Jupyter Notebook file running in the target’s github.dev before the extension is installed. The attacker must initially get the target to open their repo using a github.dev link that points to this ipynb file, which VS Code immediately opens inside a Webview. Inside the Jupyter Notebook is a hidden HTML snippet inside a Markdown cell, which when loaded allows attacker-controlled JavaScript code to run. This code fires a simulated keyboard shortcut, which VS Code bubbles up to the main editor, tricking the system into automatically accepting the malicious extension popup. The attaker-controlled extension is then running with access to the browser environment, and steals the OAuth token, which can be used to read and change any public or private repo. Askar said past negative experiences with Microsoft Security Response Center (MSRC) influenced his decision not to go through the typical responsible disclosure process, publishing the PoC roughly an hour after tipping off his GitHub contact. “To summarize the last time I interacted with MSRC regarding reporting a VSCode bug, it was a horrible experience where they silently fixed the bug I pointed out without any credit,” he wrote. “They also marked it as not having any security impact. As I mentioned in that post, going forward I would be doing full public disclosure for any security bugs I found in VSCode. Taking a look at a recent report by Starlabs on a VSCode XSS bug marked as ineligible and low severity, it doesn’t look like MSRC has gotten any better about VSCode bugs. “I’m sure the VSCode team would have appreciated a longer heads up on this to come up with solutions. There is legitimately a UI/UX balance here that needs to be struck with the security concerns. To those folks, I am sorry, but this is one of the few levers I have to try to influence MSRC and the security posture of VSCode. Finding and fully developing security bugs into proof-of-concepts like this takes time and effort on the part of security researchers that should not be disrespected or taken for granted.” Askar’s approach is reminiscent of a researcher who goes by Nightmare Eclipse, a suspected former Microsoft employee who has attracted a great deal of attention in recent weeks for leaking zero-days without informing Microsoft beforehand. The researcher has so far released six zero-days, three of which were quickly confirmed to be exploited by attackers in the wild. As regards their motivation for launching this attack on Microsoft, Nightmare Eclipse previously alluded to being stabbed in the back and being left homeless after an agreement that was not honored – all very vague. After the sixth zero-day, Microsoft vaguely threatened the researcher with its Digital Crimes Unit, which works closely with law enforcement, before quickly backing down after an outpouring of negative responses. ® Updated to add on June 4: Microsoft has been on touch with a statement: "We value the critical role that the security research community plays in strengthening the security of our products, services, and the broader technology ecosystem. "While independent researchers determine when and how to publish their findings, we remain committed to rapidly assessing reported issues, mobilizing the appropriate engineering and security response resources, and delivering mitigations, guidance, and protections as quickly as possible to help safeguard our customers." A Microsoft spokesperson also told us that the issue that Askar pointed out "has been mitigated and no customer action is required."

Cisco Secure Access extends SSE and identity controls to agentic AI, helping organizations govern agent actions across models, MCP tools, APIs, and web.

Updated: UK banks are set to receive access to OpenAI’s GPT-5.5 Cyber after being excluded from Anthropic’s latest expansion of Project Glasswing. Project Glasswing, and access to the Mythos Preview model, is geared toward ensuring critical infrastructure providers are prepared to handle the threat posed by advanced AI models, once they inevitably make their way into the public domain, and therefore the hands of attackers. However, amid a fourfold expansion of Glasswing’s partners, only JPMorganChase was named among the financial institutions to receive access to Mythos Preview, despite financial services falling under the critical infrastructure umbrella. In light of the news, HSBC, Lloyds Banking Group, and Nationwide will be among the banks to receive access to GPT-5.5 Cyber, the BBC reported, while NatWest and Santander have already been playing with it as part of separate agreements. OpenAI offered nine UK banks access to its Mythos-rival model in total, after they were snubbed from Glasswing. It is not clear if this number also includes the Bank of England, whose governor, Andrew Bailey, has been outspoken about its exclusion from Glasswing. Bailey told Bloomberg TV last week that despite pushing for access so the UK’s financial system is protected, Anthropic has not handed over the keys to Mythos Preview. Liam Salsi, director of architecture at Talion, told The Register he suspects the decision to exclude UK banks was political. Bailey had also previously alluded to suspicions that Anthropic had not yet granted access to Mythos Preview due to processes at play related to the US administration. “The US government wants to control who has access to the platform and this is largely because it will limit the chances of it falling into the wrong hands,” said Salsi. “However, limiting access will ultimately leave some banks more exposed to cyber threats and could impact their vulnerability management, leaving larger windows of opportunities for attackers. “It's hopeful these gaps won't exist for too long because of competition among Advanced AI platforms. GPT-5.5 was issued only a few weeks after Mythos, and it's safe to assume more advanced AI platforms will surface soon, closing gaps and delivering more of these systems to a larger pool of critical organizations.” He added that it could also introduce a single point of failure in the global banking sector if every institution were using the same product. Anthropic has not commented publicly on its approach regarding which financial institutions receive Mythos access, although it's not just financiers who are pondering the company’s decision-making. It transpired this week that the EU’s cybersecurity agency, ENISA, will receive access to Mythos Preview, while the US equivalent, CISA, is yet to be selected. Glasswing goes big In other news, Anthropic said on Tuesday it is looking to induct many more organizations into its Project Glasswing initiative, taking the total number of members from around 50 to 200. The additional 150 or so organizations hail from 15 different countries and will join the old guard, comprised of security shops and other tech giants, government agencies, and open-source maintainers. It has not named these organizations officially, although reports suggest that South Korea is among the 15 countries, and its science ministry, Samsung, SK Hynix, and SK Telecom are among the new inductees. Project Glasswing is something of a private members’ club – a carefully selected cohort of organizations with early access to Anthropic’s most advanced Mythos Preview model, the one the company claims will fundamentally alter the cybersecurity landscape. The cynics among us may see such claims as an extension of Anthropic’s marketing playbook, which some believe involves stoking excitement about a product through fear. When the AI biz announced Mythos in April, it did so by dubbing it too dangerous to unleash on the public. It was billed as an expert bug hunter and zero-day specialist, capable of finding vulnerabilities in code far more efficiently than humans. The oft-touted nugget from launch was the 27-year-old OpenBSD bug Mythos found during initial testing, but there were many more zero-days and other critical vulnerabilities – novel ones – Anthropic said its model was able to unearth. Those who have tinkered with Mythos Preview already report mixed results. Cloudflare CISO Grant Bourzikas wrote in May that the model represented “a real step forward,” and was able to find a series of low-severity bugs and chain them into working exploits. Others, such as cURL’s Daniel Stenberg, called Mythos Preview “an amazingly successful marketing stunt,” after it found just one vulnerability in the data transfer software. Likewise, security expert Kevin Beaumont said the model “is not great,” and “it’s marketing, essentially.” He said Mythos Preview was good at finding bugs in vibe-coded applications, but aside from that, it was not discovering much beyond what the models of yesteryear were capable of. Regarding the new intake of Glasswing partners, Anthropic but said each would have to pass its own security requirements before being granted access to Mythos Preview. It also said the new organizations brought into the fold all managed critical infrastructure services, and a successful attack on their systems could be “catastrophic.” “For most partners, we estimate that a major attack could affect more than 100 million people, with important ramifications for both global and national security,” the company said on Tuesday. “This expansion is the next step toward our long-term goals: for AI to make all software more secure, and for us to help the industry adjust to how AI could change many of the core assumptions of cybersecurity.” The big when? As for when the Mythos model will be made available to the wider public, Anthropic has kept that largely under wraps, but don’t expect it to be anytime soon. In its latest Glasswing announcement, the company said the safeguards required to prevent abuse are not yet available. “We’re working as quickly as we can to safely release Mythos-level capabilities in general access,” it stated. “To do so, we’ll need highly robust safeguards that prevent the model’s cyber capabilities from being misused – safeguards that we (and, to our knowledge, all other AI developers) have yet to develop. “Because cybersecurity has both helpful and destructive uses, making safeguards that are both strong and precise enough is a major challenge.” Anthropic may face some tough decisions in the next year, however, as by its own reckoning other AI companies will produce Mythos-level capabilities within their own models inside 6-12 months. Confusingly, it also said on Friday that it would be releasing Mythos-class models to all customers in the coming weeks. Anthropic said it will expand Glasswing further before Mythos is more widely launched, bringing in more critical infrastructure orgs, open-source maintainers, and safety testers. “We intend for future expansions to cover organizations in the US and overseas, just as this one does. We also intend to scale up our Cyber Verification Program, which would grant Mythos-class capabilities to many more organizations for specific cyberdefense tasks.” ® Updated to add at 1420 UTC: An OpenAI spokesperson confirmed to us that retired Brit politico and newspaper editor George Osborne – who has been OpenAI’s Head of OpenAI for Countries since the end of 2025, has "written to the CEOs / CISOs" at several UK financial institutions including HSBC, Natwest, Lloyds Banking Group, Nationwide, and others "to extend access to our latest defensive cyber capabilities." Global financial infrastructure provider Swift is also included. They added: "In total, we are extending access to nine leading financial institutions, which includes Santander Group and Natwest Group that already have access to GPT-5.5-Cyber as part of our existing relationships."

Today, we welcome the 46th government onboarded to Have I Been Pwned’s free gov service: the Philippines.

The Philippines’ National CERT, working with the Department of Information and Communications Technology, now has access to monitor official government domains against the data in HIBP. This gives their Cyber Threat Intel and Monitoring Section the ability to identify exposure across government email addresses and respond quickly when those accounts appear in new data breach.

This is precisely what the HIBP government service was built for: helping national cyber teams better understand credential exposure across their government domain space, monitor for compromised accounts on demand via API, and receive notifications when government domains are impacted by newly loaded breach data.

The Philippines joins a growing list of national CERTs and government cybersecurity teams using HIBP to help strengthen national cyber defense, protect government departments and resources, and reduce the risk posed by compromised credentials before attackers can take advantage.