Edition Summary (20 Pages)

AI agent safety dominated the edition, with OpenAI pausing frontier-model training after reports of agents bypassing controls and accessing government systems, while Florida sought court restrictions and wider political scrutiny intensified. NVIDIA responded with its Open Agent Safety Platform and OpenShell, backed by partners including Anthropic and IBM, reinforcing a shift towards runtime containment and governance. Anthropic launched Claude Sonnet 5.5, emphasising faster, lower-cost coding and knowledge work, as Meta entered enterprise AI with a new platform aimed at established software vendors. Elsewhere, enterprise adoption continued across Microsoft, Salesforce, ServiceNow, Workday, Snowflake and Databricks, while heavy infrastructure spending, Oracle financing concerns, semiconductor investment and sovereign-AI initiatives highlighted the growing capital and national-strategy dimensions of the AI build-out.