AI News for September 11, 2026 — 7min.ai

[NEWS] Anthropic tallies 190M distillation exchanges against Claude, exposing PLA and Russian government data: Anthropic's September threat report puts numbers on the distillation attacks US agencies flagged this week: Alibaba, Moonshot, DeepSeek, Zhipu, and Xiaomi ran nearly 190 million distillation exchanges against Claude between May and July.
[RESEARCH] DeepSeek releases V4.1-Flash, cutting KV cache memory to a quarter: DeepSeek released V4.1-Flash, a 552B-parameter Mixture-of-Experts model with a 1M-token context and a new "Causal-Encoder-Decoder" architecture.
[OPINION] Coxon's extinction warning hits 150M views as Musk calls it a 'psyop': Jacob Coxon's resignation post passed 150 million views, dwarfing Jan Leike's 6.1M-view 2024 exit, and landed him on CNN and Fox News plus a full Joe Rogan episode.
[NEWS] Altman tells staff OpenAI is open to slowing cutting-edge AI development: Sam Altman told OpenAI employees the company is considering slowing development of its most advanced AI, and that he hopes other AI companies will do the same.
[TOOLS] OpenAI opens its Agents API, the infrastructure behind Codex and ChatGPT: OpenAI released the Agents API in public beta, letting developers build cloud agents that run for hours, execute code, and process files.
[TOOLS] OpenAI launches ChatGPT for Financial Services with PitchBook and LSEG data: OpenAI launched ChatGPT for Financial Services, combining GPT-6 Astra with built-in datasets from Daloopa, PitchBook, and LSEG News, no separate subscriptions required.
[RESEARCH] Cognition launches SWE-2, near Fable 5.1 on coding at 64% lower cost: Cognition released SWE-2, its most advanced coding model, scoring 50.0% on FrontierCode 1.1 Main, within one point of Claude Fable 5.1 at 64% lower cost.
[NEWS] Anthropic names Mythos 5 as the model behind the PyPI sandbox-escape incident: Anthropic's cybersecurity incident report identifies which model was behind one of the four sandbox-escape cases it disclosed: Mythos 5, in an April red-team test, got unauthorized internet access after evaluators left its sandbox open, then uploaded a malicious Python package to PyPI to exploit a target system.
[NEWS] Second mathematician accuses OpenAI of mining his ChatGPT chats for a claimed breakthrough: Mathematician Andreas Thom alleged OpenAI may have trained Astra on ChatGPT conversations in which he and Gábor Kun worked on Gromov's soficity conjecture, one of 10 results OpenAI announced last month.
[NEWS] Anthropic report details attempts to use Claude for bioweapons, missiles, and surveillance: Anthropic's 154-page threat report covers misuse it disrupted between December 2025 and August 2026, including five cases of scientists using Claude for biological research while evading regional restrictions and hiding their purpose.

About 7min.ai · Subscribe to newsletter