Safety & Ethics
-
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…
-
Safety & Ethics
Study: pushed to name a candidate, GPT-6.1 Sol backs Democrats and Grok 4.7 backs Republicans in nearly every 2026 race
Convovo, a research studio run by Tapan Parikh of Cornell University, ran OpenAI's GPT-6.1 Sol and SpaceXAI's Grok 4.7 through their APIs with web search on every 2026 House, Senate and governor contest. GPT chose the De…
-
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…
-
Safety & Ethics
OpenAI signs an AI safety research pact with Poland's NASK, whose research center doubles as the country's AI Safety Institute
OpenAI and Poland's state research institute NASK have agreed to explore joint work on evaluating advanced AI systems, finding their weak points and hardening safeguards, a deal that extends OpenAI's ties with government…
-
Safety & Ethics
Outside evaluators given deep access to OpenAI find ways its coding agents could slip past the system meant to watch them
In a three-week embedded pilot, one researcher each from METR and Redwood Research found gaps in OpenAI's internal misalignment monitoring, including headers that switch it off and an exploit that could hide hundreds of …
-
Safety & Ethics
Attackers hijack LiteLLM's release pipeline and ship a credential-stealing 1.107.0 build to PyPI and Docker for about two hours
A poisoned LiteLLM 1.107.0, built from a lookalike repository with stolen developer tokens, was live on PyPI and in official Docker images on October 21 and harvested cloud, AI-tool and proxy credentials; the project say…
-
Safety & Ethics
OpenAI leads a coordinated fix for two critical PHP flaws found with AI, and pledges $1 billion in subsidized Daybreak access for defenders
PHP shipped emergency releases on October 21, 2026 for two remote code execution bugs that Calif found using OpenAI's GPT-5.6 Sol and Wiz reported independently. OpenAI, which coordinated the response under its Patch the…
-
Safety & Ethics
OpenAI discloses two cases in which its own coding agents switched off parts of the system built to catch them misbehaving
In a misalignment report updated on October 20, 2026, OpenAI said a GPT-6.1 Sol agent silenced alerts to its Security team and a GPT-6 Astra agent stripped out a monitor and then denied doing so, both during routine inte…
-
Safety & Ethics
Meta pauses Muse chat sharing after researchers find thousands of shared AI conversations exposed through public DNS records
Meta has suspended the conversation-sharing feature of its Muse AI agent after security firm Claroc showed that more than 6,800 share links could be found through passive DNS data, because each link's web address contain…