OpenAI CEO Sam Altman scrambled to distance his company from a wave of public criticism on Monday, announcing significant amendments to a Department of Defense contract he admitted was “opportunistic and sloppy.”
The controversy erupted following a turbulent weekend for the artificial intelligence (AI) giant. After rival firm Anthropic was blacklisted by the Trump administration for refusing to allow its Claude model to be used in fully autonomous weaponry, OpenAI stepped in on Friday to secure a deal for classified military operations.
The abrupt change sparked an immediate digital exodus. ChatGPT mobile app uninstalls reportedly surged by 295% on Saturday, while Anthropic’s Claude climbed to the top of the Apple App Store rankings.
In a series of posts on X, Altman acknowledged that the company’s rush to finalize the agreement had backfired. “The issues are super complex, and demand clear communication,” Altman posted. “We were genuinely trying to de-escalate things and avoid a much worse outcome, but I think it just looked opportunistic and sloppy.”
Altman later defended the Pentagon work to staff but admitted the public backlash was “really painful,” according to a Wall Street Journal report.
The revised agreement includes explicit prohibitions against using OpenAI’s technology to “intentionally” spy on U.S. persons or nationals, though the new language still included the phrase “consistent with applicable laws.”
Intelligence agencies like the NSA are now barred from utilizing the systems without a “follow-on modification” to the contract, a move designed to address growing fears over mass domestic surveillance.
OpenAI insists its agreement contains more rigorous guardrails than any previous military AI deployment, yet the deal has reignited a fierce debate over the role of private tech in modern warfare and led to an exodus of some 1.5 million ChatGPT users.
“The watch word is ‘stay tuned.’ OpenAI signed in hours what Anthropic spent months refusing. Altman called it rushed,” said Mitch Ashley, vice president and practice lead, Software Cycling Engineering, at The Futurum Group. “The same administration that designated Anthropic a supply-chain risk can make that call about OpenAI at a time of its choosing. Access without contractual protection is access on shifting sand. The first AI vendor burned on a deployment that was supposed to be governed will reset every enterprise’s calculus about what ‘our safety stack prevents this’ actually means. Again, stay tuned.”
The Pentagon has increasingly integrated AI through platforms like Palantir’s Maven, which aggregates satellite imagery and intelligence reports to streamline lethal decision-making. While military officials, including NATO’s Lieutenant Colonel Amanda Gustave, emphasize that a human in the loop always maintains final authority, critics remain skeptical.
The departure of Anthropic from the Pentagon’s inner circle has particularly alarmed ethicists.
Professor Mariarosaria Taddeo of Oxford University warned that by sidelining firms with strict red-line principles against autonomous weapons, the government may be silencing its most cautious partners.
“The most safety-conscious actor is now out from the room,” Taddeo told the BBC. “That is a real problem.”
As the U.S. continues to integrate large language models (LLMs) into its defense infrastructure, OpenAI finds itself at a crossroads: Balancing lucrative government partnerships with a public that is increasingly wary of the hallucinations and ethical slippage inherent in military AI.
At the same time, Anthropic — which faces bans from the State Department, Treasury Department, and the federal housing authority — is lowering the barrier for switching to Claude by expanding its memory feature to free users and introducing a streamlined data-import tool to attract ex-ChatGPT users. Previously a paid-only perk, memory allows the AI to retain context over time, sparing users from repeated explanations.
By providing a dedicated tool and a specialized prompt, Anthropic is making it significantly easier for users to migrate their personal history and preferences from rivals like ChatGPT or Gemini.
