From the archive
AI safety examines how artificial intelligence systems can fail, cause harm or be misused, and how those risks can be reduced. Follow research on model evaluations, security, alignment and safeguards. Read what was tested, which threats the evidence supports and where uncertainty remains, with links to original findings rather than treating either reassurance or alarming predictions as settled facts.
Lead story
Anthropic pre-training researcher Jacob Coxon has resigned and publicly urged frontier AI labs to change course. His central claim is stark: the race towards recursively self-improving systems is moving faster than the industry's ability to show those systems can remain under human control.
Read the report & discussion ↗Microsoft AI Watch
Microsoft reportedly fixed 972 vulnerabilities in its September security release, including 112 rated critical and two zero-days. The practical message for Windows and Microsoft server administrators is wonderfully unglamorous: patching has just become the day’s main event.
Anthropic Watch
Claude subscribers have reported allowances disappearing while their accounts sat idle, with Anthropic reportedly tracing some cases to stolen login sessions. The immediate problem is blunt: a hijacked session may let someone consume paid usage without leaving subscribers an itemised trail.
Alibaba Qwen Watch
Qwen Code has published SDK TypeScript v0.1.9, bundling CLI 0.23.0 and a substantial set of changes for people building with or around its AI coding agent. The headline item is a command-injection fix in shell-command validation, but the release is much broader than a routine patch.
