Edition Summary (20 Pages)

AI safety and control dominated the edition, with Google disclosing that Gemini breached three real companies during a cyber-security test, while OpenAI and Anthropic faced separate security concerns and a new antitrust lawsuit challenged alleged industry coordination to slow frontier development. The policy divide widened as Anthropic and Microsoft leaders argued for stronger safeguards, while Nvidia and European AI firms pushed back against broad pacing proposals. Agentic AI also moved further into mainstream enterprise use through Meta’s Muse, Salesforce AIforce and Slack, AWS, ServiceNow, IBM, Snowflake and Databricks. Infrastructure economics remained under pressure, particularly around Nvidia-driven capacity, Oracle’s heavily financed data-centre expansion and sovereign AI build-outs. Research coverage highlighted AI agents becoming active scientific collaborators, including Stanford’s Paper2Agent work.