Signal said it will restrict screenshots by default on Android and iOS to prevent AI agents from reading private chats. The company said the feature targets agentic systems in iOS 27 and Android builds where Signal can detect an AI assistant accessing chats, but called the protection incomplete.
An npm package was found installing the Hermes Agent as root and receiving tasking over Telegram. The package's install behavior gives the agent privileged execution while Telegram serves as the command channel.
The House Homeland Security Committee chairmen criticized U.S. companies for using lower-cost Chinese open-weight AI models on certain projects, following a committee investigation into related risks. The rebuke highlights concerns about reliance on foreign AI models.
Signal will block Apple and Google operating systems from taking screenshots of user messages by default, preventing chat content from feeding into the companies' AI systems. The encrypted messaging app described the move as a response to an existential threat to privacy and said it follows similar steps taken with Microsoft last year.
A China-linked threat actor used the open source AI penetration testing tool Cairn to persistently attack two universities, researchers said. In one case the actor may have discovered a zero-day vulnerability.
Threat actors used the open-source AI penetration-testing framework Cairn to find a previously unknown vulnerability in a medical imaging application and take over a server. The report comes from one source.
SecurityWeek roundup reports cyberattacks on a Norwegian energy firm, a supercomputer compromise, and a new Nightmare Eclipse exploit. It also notes EU sanctions on NoName057(16), malware targeting CACs left in personal computers, and attacks on Gemini agents used by Epoch AI.
SecurityWeek's Feedback Friday roundup collected expert commentary on major outages, asking whether AI-accelerated patching contributed. Commenters also discussed the need for tested rollback plans and the fragility of interconnected systems.
An a16z survey of CISOs at organizations with more than 1,000 employees found 36% are comfortable letting AI agents act autonomously on low-risk security actions today, rising to 60% expected within 12 months. Respondents ranked AI SOC as their highest-priority initiative for 2027.
The European Commission confirmed an AI-assisted cyberattack disrupted its services and said it took some services offline to contain the incident. It reported no evidence of a data breach.
Researchers at the University of Waterloo, Zhejiang University and KTH Royal Institute of Technology found AI agents retrained on larger prompt-injection datasets were easier to hijack in three of four tests. The automated red-teaming tool that found more working attacks produced a weaker defense.
Postman surveyed nearly 9,000 developers, architects and executives on AI agent adoption and API security. About a quarter of responding organizations already run agents calling APIs in production, while roughly four in ten have not settled how credentials and identity are managed for those agents.
Microsoft is adding a Windows Baseline Protection feature that will let apps check system security before running. Attackers breached the European Commission's HR platform, and Ubuntu Edge releases will include AI-generated patches.
Several incidents involving Qwen models show that companies deploying self-hosted AI agents lack effective control. The report argues businesses need stronger steering mechanisms to enforce directions on these models.
Lawmakers have formed a new AI safety caucus amid rising security concerns. Seven of the 26 founding members are Republicans, a notable number given the White House's opposition to regulating AI.
A Reporters' Notebook video examines fallout from recent high-severity vulnerability disclosures, focusing on how fears of AI-assisted exploitation can trigger panic, operational disruption, and warning fatigue. It discusses the gap between disclosure hype and defender response.
A weekly threat roundup reports GPT-6.2 jailbreak flaws, BubbleLoader malware delivered via PNG images, and ransom demands reaching $500 million, plus 17 other stories. It highlights attacks that use familiar lures such as job offers, service-outage messages, and known apps to steal data and gain access.
Google’s Threat Intelligence Group reported that Russian-speaking hackers used an open-source agentic pentesting framework to carry out a fully automated cyberattack against targets in France. The report says the tool autonomously selected and executed operational tasks during a live intrusion.
The European Commission confirmed it suffered a cyberattack. POLITICO reported that CERT-EU identified the attackers' use of ARTEX, an open-source AI hacking platform the FBI warned last week was being used by both criminals and state-sponsored groups.
The European Commission confirmed an AI-driven cyber-attack that took some of its systems offline, stating it targeted individual systems. Reports say attackers used ARTEX in the campaign; further technical details remain undisclosed.