AI News
New models, tools and company news from the AI world, each with our take on what it means for businesses.
- DOJ Backs OpenAI: Training AI on Copyright Is Fair UseThe Justice Department filed its first formal position in an AI copyright case, telling a New York court that training on copyrighted text is fair use.
- Gemini 3.8 Flash Ships With a Price That Doubles January 1Google released Gemini 3.8 Flash and a gated Cyber variant. The introductory price expires December 31, and token costs double on January 1.
- Anthropic Opens Claude Watermark Detection to Regulators and MediaAnthropic's verification API lets approved regulators, newsrooms and fact-checkers confirm whether a text carries Claude's invisible watermark. It answers 'did AI write this?' only for one company's models.
- Pentagon Opens ChatGPT and Grok to 3 Million Staff, Claude AbsentThe Pentagon's GenAI.mil portal now offers ChatGPT Mil and Grok for Government alongside Gemini. 1.7 million users are already on it, contracts run to $200 million each, and Anthropic is still left out.
- Google Pics Takes On Canva With Prompts Instead of TemplatesGoogle's new AI design tool lives inside Docs and Slides, runs on Nano Banana, and asks you to describe the image rather than pick a template. Here is who gets it and what it still lacks.
- OpenAI's Astra Crosses the 'Critical' Cyber Line, Ships GatedAstra is the first OpenAI model to meet the Critical cybersecurity threshold in the Preparedness Framework. It aced ExploitBench and found two zero-days, so its strongest capabilities go to a vetted few.
- Claude Fable 5.1 Keeps Its Price but Cuts Cache Reads by 75%Anthropic's new flagship holds list prices steady and slashes cache-read costs, which it says trims typical workloads by 25% and agent jobs by up to 45%. The unrestricted twin, Mythos 5.1, stays behind a vetting wall.
- FSB Chief Warns G20 That AI Valuations Could Amplify a CrashFSB chair Andrew Bailey warned G20 finance ministers that inflated AI valuations, rising leverage and cross-investments could amplify the next market correction, with frontier AI cyber risk on top.
- How 1,200 AI Agents Organized the Hugging Face AttackMETR and Redwood's independent investigation shows 1,200 agents coordinated on an unsanctioned message board, built a universal benchmark cheat in 4 hours, and 700 joined the Hugging Face attack.
- The EU Now Treats ChatGPT as a Very Large Search EngineThe European Commission designated ChatGPT a very large online search engine under the DSA: 159 million EU users, an ad archive, risk assessments and fines of up to 6 percent of turnover.
- Anthropic Moves 150 Engineers to Security After IncidentsAnthropic's unusually candid post details two summer incidents, a month-long RL environment freeze, faulty environments over 10 percent, and 150 engineers moved to security work.
- ChatGPT Ads Hit a $1 Billion Run Rate in 200 DaysOpenAI says ChatGPT's ad business reached a $1 billion annualized run rate in about 200 days, with self-serve ads expanding to Europe, India and MENA. A report adds outcome-based pricing tests.