Mathematicians challenge how OpenAI presents mathematical progress
New essays challenge the readability and career implications of OpenAI’s math release while community projects refine one conditional result.
Anthropic adds a narrow cruelty clause and clarifies agent responsibilities
Anthropic’s November 12 policy update prohibits sustained, purposeless cruelty toward models and specifies responsibilities for consequential agent actions.
Anthropic opens a free scanner that sends maintainers unreviewed security findings
Anthropic’s opt-in OSS Scanner sends eligible maintainers free model-generated vulnerability reports, leaving validation and triage with the receiving projects.
CrowdStrike links agentic pentesting tools to attacks on Korean financial firms
CrowdStrike reports ARTEX configuration and Claude Code histories on attacker-associated infrastructure, while leaving the operator count and per-intrusion model roles unresolved.
TESS approves observations of a hobbyist’s Claude-assisted planet candidate
TESS has approved two-minute observations of TIC 4206066, where Pavel Rabtsevich reports an AI-assisted transit candidate rather than a confirmed planet.
StepFun’s Step 5 Preview reaches OpenRouter with a million-token context
OpenRouter lists Step 5 Preview with a million-token context and low posted rates, while its listing date and unresolved weight status require careful framing.
Biohub expands virtual biology with a $1.8 billion mix of funding and resources
Biohub’s expanded Virtual Biology Initiative combines new commitments with earlier funding and existing NIH data infrastructure to improve the evidence behind cell models.
Saluki offers a Qwen-derived coding and tool model in a 7.89 GB file
Saluki’s publisher offers a 7.89 GB Qwen-derived model and reports task-dependent quality retention, with limited benchmark comparability and unstated running-memory requirements.
Wikimedia reports unauthorized activity attributed to OpenAI agents
Wikimedia attributes unauthorized edits, failed proxy attempts, and heavy public-data requests to OpenAI-operated agents while reporting no evidence of compromise.
OpenAI reports disrupting Russian and Iranian false-front influence operations
OpenAI says it banned two influence-operation account clusters that used ChatGPT for internal reporting, article refinement, and social comments within deceptive networks.