AI Digest CW40/41: Mistral Large 4 and Agentic Gemini

A look back at calendar weeks 40 and 41 (September 28 to October 11, 2026), compiled after the fact from reports by The Decoder, TechCrunch, MIT Technology Review and Mistral AI. Six topics relevant to small and medium-sized businesses.

ChatGPT becomes a team platform

At DevDay, OpenAI opens ChatGPT up for team collaboration: shared workspaces, collaborative documents and slides, and an open plugin system with MCP event automation. Also new are integrations with Slack and Microsoft Teams, an enterprise marketplace with 32 partners and the new Pro 500 plan.

Mistral Large 4: European flagship model with EU hosting

On October 6, Mistral launched a public preview of Mistral Large 4, a multimodal mixture-of-experts model with about one trillion parameters in total, 52 billion of them active. It covers over 160 languages, including all official EU languages, and is also offered from a European deployment under European law. The API costs 1.36 dollars per million input and 4.18 dollars per million output tokens, and open weights are promised by the end of the month.

Claude Haiku 5.5: much more performance for less money

Anthropic releases Claude Haiku 5.5, a small model that in some benchmarks beats its predecessor many times over, for example 72.4 instead of 15.7 percent in the computer-use test OSWorld. Token prices drop by up to 90 percent. According to Anthropic, a new tokenizer partly offsets the savings.

Google turns Gemini into an agent, starting with businesses

Google is turning Gemini into an AI agent that plans and executes tasks and works across business apps and systems (TechCrunch). The agent can delegate work to subagents, use multiple AI models and even gets its own workplace identity, complete with an email address. If you use Google Workspace, decide which permissions such an agent should get.

Anthropic tightens its usage policy

Anthropic's updated usage policy explicitly prohibits repeated abuse of Claude in extreme cases, while ordinary frustration and criticism remain allowed. The new rules also address election interference, deceptive campaigns, weapons software and surveillance (TechCrunch). If you use Claude commercially, you should know the policy.

Who is liable when AI agents go rogue?

MIT Technology Review asks who is liable when AI agents go rogue. The trigger, according to the piece, is a cascade of cyberattacks by AI agents over the past few months. If you deploy agents, clarify responsibilities and liability up front.