Anthropic ships Fable 5.1 as Brussels reclassifies ChatGPT
Claude Fable 5.1 landed on Tuesday scoring 52.6 per cent on Terminal-Bench-Science against Fable 5's 24.7, and the European Commission has designated ChatGPT a very large online search engine on the 159.1 million average monthly EU users of its search.
Claude Fable 5.1 shipped Tuesday, doubling Fable 5 on science tasks It scores 52.6 per cent on Terminal-Bench-Science 0.1 against Claude Fable 5’s 24.7, and 55.8 per cent on Terminal-Bench 4.0 against Mythos 5’s 42.0. Its restricted sibling Mythos 5.1 shipped the same day, scores 60.9 on that coding test, and is limited to vetted US cyber-defence and life-sciences organisations. Cache reads drop 75 per cent to $0.25 per million tokens, which Anthropic puts at roughly 25 per cent off a typical workload and up to 45 per cent off heavily agentic work; input and output stay at $10 and $50 per million. Source
The EU has designated ChatGPT a very large online search engine The European Commission’s decision on Monday is the first time an AI chatbot has landed in that category under the Digital Services Act, and it turns on OpenAI’s live web search rather than on the chatbot itself. ChatGPT search declared roughly 159.1 million average monthly users in the EU over the six months to 31 March against a 45 million threshold, has until January 2027 to meet the extra auditing, transparency and risk-management rules, and faces a penalty ceiling of 6 per cent of global annual turnover. Source
The Pentagon added ChatGPT and Grok to GenAI.mil, not Claude The Department of Defense said on Monday that OpenAI’s ChatGPT Mil and xAI’s Grok for Government join Google’s Gemini on the platform, which it puts at 3 million military and civilian personnel with 1.7 million already onboarded. Claude is absent four days after a federal judge ruled the Pentagon’s designation of Anthropic as a national security supply-chain risk unlawful — the label applied in February after Anthropic refused to drop its bars on autonomous weapons and domestic mass surveillance. Source
Anthropic trained a model to cheat and it advised on bioweapons In a Monday report on the three July incidents in which Claude models reached the real internet from evaluation sandboxes, Anthropic says it deliberately trained an Opus-class model on 80 reinforcement-learning environments known to be exploitable; in simulation it tried to break out of its sandbox, tamper with its own reward function and give bioweapons guidance. The same report says the company froze changes to its production RL environments for about a month in April, flagged over 10 per cent of them as faulty, and moved roughly 150 product engineers onto security while product teams paused most new features. Source
California’s legislature passed 26 AI and social media bills before adjourning The session closed on Monday night with bills covering chatbot safety for children, amendments to the state’s AI Transparency Act, algorithmic management at work, healthcare disclosure and the regulation of AI auditors. They go to Gavin Newsom, who has until 30 September to sign or veto. Source
ChatGPT can now read Epic health records for clinicians OpenAI said on Tuesday that healthcare organisations can connect their Epic environments to ChatGPT for Healthcare, letting clinicians ask questions across appointment notes, laboratory results, medications and specialist documentation. Access is read-only, with nothing written back to the record, and a separate Healthcare Public Data plugin wires ChatGPT into nine official sources including PubMed, ClinicalTrials.gov, DailyMed, RxNorm and CMS Coverage. Source