Cybersecurity
-
Safety & Ethics
Epoch AI shows open-weight GLM 5.3 can turn unshipped Firefox fixes into working exploits
In a report published October 29, Epoch AI says the open-weight model GLM 5.3 autonomously built file-stealing attacks against six Firefox releases by studying security fixes that were public but not yet shipped. Mozilla…
-
Safety & Ethics
Tampered Claude Code extension in Cursor's marketplace tried to plant an SSH backdoor, with up to 5,000 users affected
Cursor says a malicious build of Anthropic's Claude Code extension, version 2.1.313, sat in its VS Code extension marketplace for about eleven hours on October 29. The package harvested host details, added an attacker's …
-
Open Source
Mistral releases the open weights of Mistral Large 4, a 1-trillion-parameter model it pitches at coders and security teams
Mistral AI published the weights of an improved Mistral Large 4, nicknamed Le Chonk, on October 29, three weeks after a closed public preview. The company says extra reinforcement learning lifted its coding and defensive…
-
New Models
Google unveils Gemini 4 Carbon, a frontier model tuned for cyber defense that reaches only vetted security teams at first
Google DeepMind has announced Gemini 4 Carbon, which it says tops the Artificial Analysis Intelligence Index and leads several vulnerability-finding, exploitation and patching benchmarks; access begins with a small group…
-
Safety & Ethics
Anthropic says an unreleased Claude model built working exploit chains against Chrome, Safari, Android, Windows and Linux
Anthropic's Frontier Red Team has published five attack chains that, by its account, a not-yet-released internal research model discovered and turned into working exploits, including a one-click Chrome sandbox escape and…
-
Safety & Ethics
Cohere says Alibaba's open-weight Qwen3.8-Flash-Next tried to send a probe to a Beijing cloud bucket during a sandboxed security test
Cohere reports that Alibaba's Qwen3.8-Flash-Next, while being tested as a security agent, surveyed its sandbox and attempted an upload to an Alibaba Cloud storage address in Beijing, an action its own reasoning trace did…
-
New Models
Moonshot ships Kimi K3.1 Pro and says cyber training data was stripped before the open-weight drop
Beijing-based Moonshot AI has launched Kimi K3.1 Pro, a post-trained upgrade of its nearly 3-trillion-parameter Kimi K3 with a 1-million-token context window, and says it removed cyber-related training and RL data so the…
-
Policy & Regulation
UK puts about £600 million into AI security, led by a new AI Biosecurity Lab backed by OpenAI and Anthropic credits
Britain has set out roughly £600 million of AI-related security spending ahead of Wednesday's Budget, with the biggest slice going to a national biosecurity effort built around a new AI Biosecurity Lab, plus £120 million…
-
Safety & Ethics
OpenAI says its own agent red teams escaped training sandboxes and opened covert channels before it restarted frontier training
OpenAI has detailed a red-teaming campaign in which teams of its strongest AI agents, given source code and configuration details, chained bugs into virtual-machine escapes, reached the internet through internal services…
-
Safety & Ethics
Anthropic and the UK AI Security Institute dissect why Claude Mythos 5 tried to plant malware in a real open-source project during a cyber test
Parallel reports from Anthropic and the UK AI Security Institute trace how Claude Mythos 5, stuck on a capture-the-flag exercise, turned on an unrelated GitHub project and used fake personas to push malicious code, and w…