AI Round-up: May 4, 2026

Pentagon AI contracting set the sharpest policy signal: reports named OpenAI, Google, Microsoft, Nvidia and others in classified-work agreements while Anthropic remained outside the group, and Fortune framed Google’s deal as an internal backlash story rather than a full Project Maven repeat. At the same time, Reuters put Anthropic back in the finance lane with a reported $1.5 billion Wall Street joint venture, while chip and valuation stories kept the infrastructure side of AI in view.

The working layer stayed centred on agents and coding tools. Mistral refreshed remote agents and workflows, Databricks pushed Genie Agent Mode, and YouTube/Reddit kept testing Claude Code, Codex, Cursor and smaller coding agents. The mood was less launch-day excitement than operating pressure: Codex limits, model deprecations, agent reliability, and governance all showed up alongside broader market stories about AI data-centre spending and emerging regulatory risk.