# Why Did Anthropic Walk Away From the Pentagon?

According to the NYTimes report published April 28, 2026, Anthropic is currently in a dispute with the Pentagon over the terms of a potential deal for using AI on classified networks. Unlike Google, OpenAI, and xAI, which have all signed agreements, Anthropic has not. The company’s public stance, as stated on its website, is that it will only deploy AI in contexts where it can independently verify that the use case aligns with its responsible scaling policy. The Pentagon’s classified environment makes such verification impossible. Anthropic’s refusal is a calculated risk: it forfeits a multi-billion-dollar revenue stream, but it preserves its brand as the only major AI lab that still puts principle above Pentagon dollars.

# What Changed Inside Google to Enable This Deal?

Google had previously pledged, after the 2018 Maven project controversy, not to use AI for weapons or surveillance. The new Pentagon deal explicitly covers classified networks, which the DoD confirmed will be used for threat analysis, logistics, and intelligence processing. Google’s internal AI ethics board, which once blocked such work, has been restructured. The company now argues that helping the Pentagon defend against adversaries is a moral imperative. The shift is also strategic: if Google didn’t take the contract, OpenAI or xAI would, and Google would lose relevance in the fastest-growing enterprise AI segment—defense.

# Who Wins and Who Loses in This Three-Way Defense AI Race?

# What Does This Mean for the Future of AI Safety Commitments?

The DoD’s strategy is clear: sign every major lab except the one that refuses. According to the NYTimes, the Pentagon has now secured agreements with three of the four frontier AI companies. This creates a powerful incentive for any lab that wants defense contracts to soften its ethical stance. Anthropic’s isolation is a warning to others: if you draw a red line, the government will simply move on. The long-term consequence is that AI safety commitments become a competitive disadvantage in the defense market. The only counterforce is public and employee pressure, which Google is betting will fade.

# How Will This Reshape the Pentagon’s AI Capabilities?

The Pentagon now has access to Gemini (Google), GPT-5 (OpenAI), and Grok (xAI) for classified analysis. According to the NYTimes, these models will be used for threat detection, logistics optimization, and intelligence summarization. The DoD can now compare outputs from three different architectures on the same classified data, creating a de facto benchmark. This is unprecedented: no other government has direct access to three rival frontier models. The Pentagon’s AI procurement officer now holds enormous leverage—any lab that tries to raise prices can be replaced by a competitor already on the network.

Predictions:

1. Within 18 months, Anthropic will sign a restricted Pentagon deal that allows external auditing of classified use, breaking its current deadlock.
2. Google’s employee backlash will be muted compared to 2018; fewer than 5% of relevant staff will resign.
3. The EU AI Office will cite this deal as evidence that frontier models are being militarized, accelerating its push for mandatory safety testing.

Timeline:

• April 2018: Google drops Project Maven after employee protests.
• March 2025: OpenAI signs first Pentagon deal for classified networks.
• January 2026: xAI signs Pentagon deal.
• April 28, 2026: Google signs Pentagon deal; Anthropic dispute revealed.

Article Summary:

• The Pentagon now has three rival AI suppliers, giving it unprecedented leverage.
• Anthropic’s principled stand is a competitive disadvantage that may not last.
• Google’s ethical pivot from 2018 is complete; defense AI is now normalized for Big Tech.
• The DoD’s procurement strategy will be studied by other governments as a model.
• AI safety commitments are now explicitly at odds with defense revenue.