Anthropic Investigates Elevated Errors Across Five Claude Models

📢 A stock research firm used Claude to design proteins and sent them to a real lab. A financial research firm spent $200,000 testing whether AI could design entirely new proteins that actually work in a laboratory. The five-person team included three people with no scientific background. They used Anthropic’s more advanced internal Claude model, running it on the company’s own machines. In a single night, Claude reportedly launched more than 300 sub-agents across 300+ GPUs to design and evaluate proteins. The team initially hit biosafety roadblocks, including a questionnaire with more than 60 questions about safety procedures they didn’t have in place. They eventually focused on scorpion venom, aiming to explore a different problem from the snake-venom research already underway at the Baker laboratory. The experiment highlights a strange reality: AI is making sophisticated biological research accessible to people who aren’t trained scientists, but physical experiments, safety requirements and laboratory access remain real obstacles. Source. @aipost 🏴

Monitoring - We are seeing success rates return to normal for all affected models, and are continuing to monitor to ensure no further issues. Oct 23, 23:54 UTC Identified - We have identified the cause of elevated errors on requests to Claude Fable 5.1, Sonnet 5.5, and Haiku 5.5, as well as fast mode requests to Opus 5.5 and are working on a fix. We will provide an update as soon as possible. Oct 23, 23:39 UTC Investigating - We are investigating elevated errors on requests to Claude Fable 5.1. We will provide an update as soon as possible. Oct 23, 23:37 UTC Post flair and post body will be updated as the incident report is updated by Anthropic. This discussion post will be removed from subreddit highlights one hour after the incident is resolved. View this incident on status.claude.com su

A 24-hour evaluation of six major AI personal agents on an iPhone tested proactivity, direction following, and browser control across five real-world scenarios including news briefings, schedule conflicts, and travel planning. The tested systems included offerings from OpenAI, Meta, Google, Anthropic, and xAI, with varying price points and interface models. https:// en.hacks.gr/dokimi-se-iphone-p os-ta-pigan-exi-psifiakoi-voithoi-technitis-noimosynis-se-24-ores/ # ArtificialIntelligence # Apple # OpenAI # Google