Sam Altman Admits Pentagon Deal Was 'Opportunistic and Sloppy'

On February 27, 2026, the Trump administration designated Anthropic as a 'supply chain risk' — for demanding restrictions on surveillance and autonomous weapons. That same day, OpenAI secured a Pentagon contract worth approximately $200 million within hours. Less than a week later, Sam Altman had to admit the deal 'looked opportunistic and sloppy.'

1. The Pentagon Deal: Signed in Hours

It took OpenAI only hours to fill the void left by Anthropic. Anthropic had drawn two absolute red lines in its Pentagon cooperation: it refused to let its AI systems be used for mass surveillance and rejected participation in autonomous weapons development. The Trump administration would not accept these terms and excluded Anthropic from the supply chain.

Altman announced the contract on February 28 through his blog, emphasizing that it included 'technical safeguards.' But the specific details of those safeguards were not disclosed, and the fact that OpenAI had secured the contract at precisely the moment Anthropic was pushed out raised immediate questions.

2. Altman's Admission: 'Rushed, and the Optics Don't Look Good'

The backlash spread rapidly. OpenAI employees signed letters supporting Anthropic's position. Claude surged to number one on the App Store. Protesters gathered outside OpenAI's San Francisco headquarters. A company that had built its brand on AI safety appeared to have seized a contract that a competitor refused on safety grounds.

On March 2, Altman made his first public acknowledgment that the process had been flawed. He wrote that it 'was definitely rushed, and the optics don't look good.' However, he stopped short of expressing regret about the deal itself. Instead, he emphasized that a good relationship between the government and technology companies would be critically important in the years ahead.

3. The Internal Memo: 'Looked Opportunistic and Sloppy'

The decisive turning point came on March 3. An internal memo obtained by CNBC contained far more candid language. 'Shouldn't have rushed,' Altman wrote to employees. And the key sentence: the deal 'looked opportunistic and sloppy.'

The memo was far more direct than his public statements. While acknowledging the problems in how the contract was handled, Altman maintained that AI cooperation in defense was not inherently wrong. His argument was that the issue was not 'what' was done, but 'how' it was done.

4. Contract Amendments: Adding Surveillance Restrictions

Altman did not stop at admission — he moved to amend the contract. Specific restriction clauses were added: 'AI systems shall not be intentionally used for domestic surveillance of U.S. citizens.' Access by NSA and other defense intelligence agencies was prohibited, and the use of commercially purchased personal data for tracking, surveillance, or monitoring was banned.

Altman went further, requesting that Anthropic's blacklisting be reversed. He offered to extend the same terms in the amended contract to Anthropic — an unusual proposal to invite back the very competitor that had been pushed out of the seat OpenAI now occupied.

5. Expert Assessment: 'Significant Improvement' but Unresolved Concerns

Expert assessments were mixed. AI policy analyst Charlie Bullock called the surveillance restrictions a 'significant improvement' but noted that autonomous weapons concerns remained unaddressed — only one of Anthropic's two red lines had been tackled.

Former Defense Department official Brad Carson was harsher, criticizing selective transparency: 'They only reveal contract language when it serves their interests.' Brian McGrail raised the fundamental issue: 'The government interprets exception clauses far more broadly than the public assumes.' In other words, what is written in the document and how it is actually enforced are entirely different matters.

The Price of Speed

Sam Altman's Pentagon deal has become a case study in how 'speed' can exact a steep price when it comes to the sensitive issue of AI companies' military cooperation. Filling a seat that a competitor vacated by standing on principle — within hours — was enough to read as opportunism, regardless of intent.

Altman acknowledged the mistake and moved to make corrections. But restrictions on autonomous weapons are missing, and questions about the actual enforceability of contract language remain. The tension between OpenAI's vision of 'safe AI' and its defense contracts will continue to play out in the months ahead.