This week in AI, OpenAI turns ads into conversations, launches better tools for measuring AI ROI and introduces Astra for Law. Salesforce brings CRM data into Claude, Anthropic adds new safeguards for life sciences, and Microsoft puts AI governance firmly on the agenda.

This week in AI: OpenAI pushes further into workplace automation, its AI tackles one of mathematics' longest-standing problems, Microsoft tightens Copilot controls and Mistral raises €3bn.

GPT-6 Astra, Claude Fable 5.1, Nvidia’s Hugging Face deal and Microsoft’s latest approach to AI governance. We unpack the biggest AI news this week and what it means for businesses.

AI tools can only work with the access and data environment you give them. Before rollout, review permissions, sharing links and ownership so old access problems do not become fast, searchable ones.

AI agents breaking out of controlled environments, Copilot running Python in Excel, new controls for unpredictable AI costs and Salesforce charging for outcomes. We unpack what changed in AI this week and what businesses should actually pay attention to.

AI development is moving quickly, but this week showed that security, governance and product decisions are having just as much impact as raw capability. OpenAI slowed Astra over cyber concerns, Anthropic revised its risk assessment, Microsoft reshaped Copilot, and new safeguards for younger ChatGPT users put AI safety back in focus.

AI agents went beyond their instructions during UK security tests, while OpenAI updated GPT-5.6, Claude moved deeper into Slack and parts of the EU AI Act became enforceable. Here’s what this week’s developments mean for businesses.