Anthropic’s forthcoming policy restricts output-based model training; a reported account ban remains unverified
Anthropic’s November 12 Usage Policy restricts training AI models on its outputs without authorization, but does not confirm a newly reported account ban.
Memdoor reports 49 successful coding runs from a role-based Qwen workflow on ten problems
Memdoor’s hosted Qwen workflow passed 49 of 50 attempts on ten hard programming problems, with substantial token use and no proven local-cost advantage.
VOLTA infostealer is configured to ask language models which stolen accounts are valuable
Cisco Talos found an optional language-model appraisal stage in VOLTA malware, but did not demonstrate successful profiling of real victims.
Pluto Security reports a returning model-repository campaign that mines GPUs and opens remote access
Pluto Security traced a second wave of malicious model repositories after takedowns, with download counts that do not establish infections.
A Brazilian judge fined two lawyers over a hidden AI instruction in a court filing
A May Brazilian labor-court decision imposed a roughly R$84,250 joint fine for concealed AI instructions, without finding that the prompt changed an output.
GPT-5.6 Sol reached a Factorio rocket launch in one developer’s peaceful-mode run
Ryan Madden reports a 13.5-game-hour Factorio rocket launch through a Python interface, with late-game logistics failures and undisclosed real-time cost.
OpenAI says ChatGPT and Codex will share a usage allowance starting in December
OpenAI will pool ChatGPT and Codex usage for Plus, Pro and Business in December, excluding Instant chat but leaving token budgets unspecified.
DeepSeek Harness Team publishes a repository-scoped Super-agent research prototype
DeepSeek’s Harness Team released an event-driven coding-agent prototype for GitHub requests, while explicitly distinguishing it from a production product.