Early March 2026. Frontier AI Labs in Turmoil, Military Access Clash, Model Momentum, and Outages

Ronni Holmvig Strøm · 2026-03-02

The past two weeks have delivered one of the most dramatic episodes in frontier AI governance.

A high-stakes standoff between the U.S. Department of War (formerly Defense) and Anthropic over AI usage restrictions escalated into bans, blacklisting threats, employee solidarity letters, and a swift pivot by OpenAI to secure a similar deal.

Meanwhile, model releases and massive funding rounds continued unabated, underscoring the relentless pace of capability advancement amid deepening ethical and geopolitical fractures.

Anthropic: Principled Resistance Meets Government Backlash

Anthropic's refusal to grant unrestricted military access to Claude models triggered swift repercussions.

CEO Dario Amodei publicly reaffirmed the company's red lines: no use for mass domestic surveillance of U.S. citizens and no deployment in fully autonomous weapons systems without meaningful human oversight. Anthropic highlighted its pioneering role—first frontier lab to deploy on classified networks, at national labs, and with custom models for national security—but rejected demands to remove these safeguards.

The Pentagon responded aggressively: President Trump directed all federal agencies to cease using Anthropic technology (with a six-month transition), and Defense Secretary Pete Hegseth designated Anthropic a "supply chain risk", a label typically reserved for foreign adversaries—barring contractors from engagement.

Agencies including Treasury, Federal Housing Finance Agency, Fannie Mae, and Freddie Mac began terminating Claude usage.

Tech workers rallied, Over 300 Google DeepMind employees and 60+ from OpenAI signed an open letter ("We Will Not Be Divided") urging their companies to adopt Anthropic's red lines and stand united against Pentagon pressure.

Consumer sentiment swung in Anthropic's favor: Claude briefly topped the Apple App Store, fueled by boycotts of OpenAI over its subsequent deal.

On March 2, Claude suffered a widespread outage amid "unprecedented demand" (free users +60% since January; paid subscribers doubled since October). Consumer-facing services (claude.ai, apps, Opus 4.6, Console, Code) were hit, though enterprise/API remained stable. Service restored later that day.

Product momentum persisted quietly. Claude Opus 4.6 and Sonnet 4.6 (February releases) delivered strong red-teaming and cybersecurity results, including vulnerability discovery in open-source code.