What Happened
The Economist published a briefing on March 5 arguing that the escalating conflict between the US government and Anthropic is making an AI disaster more likely. The piece centers on Defense Secretary Pete Hegseth's designation of Anthropic as a "supply-chain risk" after the company declined to sign a Pentagon AI deal, while OpenAI agreed to provide services to the Department of Defense.
The dispute has gotten personal. A leaked internal memo showed Anthropic CEO Dario Amodei calling OpenAI's public statements "mendacious" and "safety theater," describing Sam Altman's comments as "straight up lies" and "gaslighting." Meanwhile, the US military deployed Maven Smart System - built by Palantir and incorporating Anthropic's Claude model - to help identify targets during operations in Iran, accelerating decision-making from weeks to near-real-time.
Both companies have weakened their own safety commitments under competitive pressure. Anthropic revised its Responsible Scaling Policy, dropping its commitment to withhold model development until safety measures exist. OpenAI removed explicit bans on military applications from its policies back in 2024.
Why It Matters
If you're a Claude user, today nothing changes about your API access or product experience. But the structural dynamics here matter for everyone who depends on AI tools.
The AI safety ecosystem has been built largely on corporate self-regulation. Anthropic positioned itself as the company that would slow down when necessary. OpenAI claimed to prioritize safety over profit. Both have now compromised those positions. With minimal government regulation and stalled international AI safety efforts, there's no external check on development speed.
The concentration problem is real. A small number of corporate leaders and government officials are making decisions about how these systems get built and deployed. When those relationships break down, the informal guardrails that existed through cooperation break down with them.
For teams choosing between AI providers for sensitive work, the political dimension is now a factor. Government-adjacent organizations face real procurement risk with Anthropic while the supply-chain label persists. Private sector users are unaffected practically, but the funding and strategic implications ripple outward.
Our Take
The Economist piece points at something uncomfortable: the "responsible AI" era might already be over, and competitive pressure killed it before regulation could catch up.
Anthropic's whole pitch was that they would be the safety-first lab. That positioning attracted talent, funding, and trust. But when the Pentagon comes calling with billions in contracts and your competitor says yes while you say no, the market punishes restraint. And when your own models end up in military targeting systems through third-party integrations anyway, the moral high ground gets harder to claim.
What's missing from this picture is any credible external governance. The US has no comprehensive AI regulation. International efforts have stalled. The result is that two companies having a public feud is the closest thing we have to a policy debate about AI in defense applications.
For Claude users and AI practitioners: pay attention to the structural dynamics, don't panic about the headlines. The tools still work. But the institutional framework around how they're developed and governed is weaker than it looked a year ago.