Anthropic CEO Slams OpenAI's Pentagon Deal as 'Straight Up Lies'

Anthropic CEO Dario Amodei has accused OpenAI of lying about the safety protocols in its new military contract with the Department of Defense. While Anthropic rejected the deal over concerns about mass surveillance and autonomous weapons, OpenAI claims its contract forbids such uses under current law. This dispute has led to a massive user migration, with ChatGPT uninstalls skyrocketing while Anthropic's Claude app reached record-high rankings.

Key Points

Anthropic abandoned a $200 million Pentagon deal because the military would not rule out using AI for autonomous weapons or domestic surveillance.
OpenAI signed a deal with the Department of Defense shortly after, claiming to have secured the safety safeguards Anthropic could not.
Dario Amodei accused OpenAI of 'safety theater,' arguing that their 'lawful use' protections are a loophole since laws can be changed by the administration.
Public backlash against OpenAI's military involvement led to a 295% spike in ChatGPT uninstalls and a surge in popularity for Anthropic's Claude.
Nvidia is ending its investment cycle in both companies, citing their upcoming IPOs and the increasingly complicated competitive and ethical landscape.

Sentiment

The community overwhelmingly sides with Anthropic and against OpenAI on this issue. Most commenters view Sam Altman's claim of equivalent safeguards as transparently deceptive, and the 'all lawful use' framing is widely mocked as meaningless given the government's ability to redefine legality. While there is a vocal minority questioning Anthropic's sincerity given the Palantir partnership and the strategic benefits of their stance, the dominant sentiment treats Amodei's position as principled and Altman's as dishonest. Political discussions about surveillance, government overreach, and corruption run throughout, reinforcing the pro-Anthropic consensus.

In Agreement

OpenAI's 'all lawful use' clauses are effectively meaningless since the government defines what is lawful, including through secret FISA court decisions, making it a blank check for the state
Anthropic demonstrated genuine integrity by walking away from a $200M contract rather than engaging in 'safety theater' — the DoD rejected their red lines and then accepted OpenAI's weaker terms, proving the conditions were substantively different
Greg Brockman's $25M donation to a Trump-supporting PAC alongside the contract award looks like straightforward corruption and pay-to-play politics
The government changed the terms after initially agreeing to Anthropic's restrictions, then punished Anthropic for holding firm — showing the real issue was compliance, not safety
AI dramatically enhances surveillance capabilities by enabling autonomous analysis and judgment of massive datasets that previously required enormous manual effort, making Anthropic's red lines on mass surveillance genuinely important
Anthropic's stance will attract top AI talent who care about safety and don't want their work weaponized without guardrails

Opposed

Anthropic's partnership with Palantir, a surveillance-focused company, undermines their moral authority on these issues — they were already enabling the very things they now claim to oppose
Companies shouldn't have the ability to dictate how the military uses its tools, as this gives private entities veto power over democratically-elected government decisions
App Store rankings and uninstall surges are fleeting metrics that don't represent meaningful market shifts — similar to Lyft briefly overtaking Uber
Anthropic may be engaged in calculated strategic positioning for talent and market differentiation rather than acting from genuine principle
AI doesn't fundamentally change surveillance capabilities that already exist through cloud computing and big data systems — it merely acts as a better interface
The enforcement problem makes Anthropic's conditions impractical regardless: auditing classified logs poses security risks, while hard-coding ethical limits creates an unreliable AI kill switch