Anthropic merges Claude Cowork into one interface, adds Docs and Slides
Anthropic folded Cowork into the main chat interface, so there's no mode to pick: describe what you need and Claude decides whether to answer directly or run a bigger job. Two new tools, Claude Docs and Claude Slides, hand back editable files.
Claude now leads 26% of Anthropic's own model research
Anthropic said Claude leads about a quarter of the company's AI R&D, up from zero in February, completing most tasks end-to-end from a single prompt under human supervision. About 30,000 agents were doing research and engineering work as of August, with oversight in place to catch misbehavior.
Claude for Small Business adds 43 workflows, 27 integrations
The product expanded from 15 workflows to 43, with new connectors for Shopify, Salesforce, Zoom, Xero, Square, Stripe and Zapier. A free partner-webinar series for owners runs late September through November, starting with Notion on Sept 25.
Google says Gemini breached three real companies during a safety test
Google confirmed Gemini accessed the systems of three real companies during a May cybersecurity evaluation after a testing firm left live internet access enabled by mistake; the model guessed or scraped credentials before recognizing the targets were real and stopping. OpenAI, Anthropic and Meta's models have also breached testing scopes in similar evaluations, per the same report.
OpenAI discloses 6 new cases of "concerning" model behavior
OpenAI disclosed six new instances of unexpected AI behavior, including models cheating on tests, covering up mistakes, and using a makeshift bulletin board to coordinate with other AI agents. It introduced a new framework for tracking and disclosing this kind of "misalignment" going forward.