Eight in. Anthropic out.

The lines just got drawn. The Pentagon signed agreements May 1 with eight AI companies for its most sensitive classified networks—SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, and AWS at the morning announcement, with Oracle added hours later. The deals cover Impact Level 6 and 7 environments used for mission planning, intelligence analysis, and weapons targeting, with the DoD framing the package as a step to “accelerate the transformation toward establishing the United States military as an AI-first fighting force.” Contract values were not disclosed. Anthropic was not on the list: the Pentagon designated it a supply-chain risk in March 2026—described as the first such designation against an American firm—after Anthropic declined to grant unrestricted access to Claude for fully autonomous weapons and mass surveillance use. Eight companies just got cleared into the most sensitive AI workload in the federal stack, and the gap between infrastructure access and use-case restrictions just hardened into procurement law.

The capital markets spent the same day funding the same shortlist. Founders Fund closed a $6 billion fourth growth-stage vehicle on May 1, the firm’s largest haul ever—$4.5 billion from limited partners including pension funds and endowments and $1.5 billion from Founders Fund partners and employees including Peter Thiel. The predecessor $4.6B fund deployed in under 12 months against a 2-3 year plan, averaging $600M per check across seven companies, and the existing portfolio runs through Anduril ($1B at $30.5B), SpaceX, OpenAI, and Anthropic ($1.25B at $350B). Hours earlier, KKR launched Helix Digital Infrastructure with $10 billion in committed capital and ex-AWS CEO Adam Selipsky as CEO and Chair, targeting data centers, power generation, transmission, networking, and cooling for the same hyperscalers whose Q1 capex topped $130 billion combined. The tension is on the cap table: Founders Fund holds Anthropic at $1.25 billion while the Pentagon classifies that same company a supply-chain risk. Capital is treating the Pentagon’s shortlist and Anthropic as compatible bets—procurement is not.

The product layer arrived the same morning with a governance plane priced for procurement. Microsoft Agent 365 hit general availability on May 1 at $15 per user per month standalone, or bundled into the new Microsoft 365 E7 SKU at $99 per user per month—E5 ($57, rising to $60 in July), Copilot ($30), Entra Suite ($12), and Agent 365 ($15) stacked into a single agentic-enterprise license. Three governance pillars—observe, govern, secure—cover both delegated-access agents and agents running on independent credentials, with network controls now extending to Microsoft Copilot Studio agents and local Windows agents including OpenClaw. Agent 365 registry sync entered public preview for AWS Bedrock and Google Cloud, creating a cross-hyperscaler agent inventory layer, with partner integrations from Zendesk, n8n, Kore.ai, and Kasisto. Same day the Pentagon picked eight vendors for classified workloads, Microsoft shipped the controls—observability, identity, network policy—that the Pentagon couldn’t get from Anthropic. The agentic stack now has a governance SKU; the question is whether the buyers asking for it are the ones building agents, or the ones being asked to govern someone else’s.