Sam Altman delivered a blunt message to OpenAI employees during an all-hands meeting on Tuesday: the company doesn’t get to make operational decisions about how the US military uses its artificial intelligence technology.

The OpenAI chief executive told staff that whilst the Pentagon values the company’s technical expertise and will allow it to build safety protections, ultimate decisions about military applications rest with Defence Secretary Pete Hegseth.

The meeting, held four days after OpenAI announced a deal allowing the Department of Defence to deploy its models on classified networks, comes amid significant internal and external backlash over the arrangement.

WATCH MARKETING EDGE ONTV

According to people familiar with the matter, Altman used hypothetical military actions to illustrate the boundaries of OpenAI’s role. He told employees they don’t get to weigh in on whether specific military operations are justified or not.

The Pentagon has made clear it respects OpenAI’s technical expertise and wants input about where the models fit well. The government will allow the company to build what Altman called a “safety stack” to prevent misuse. If a model refuses to perform a task, the military won’t force OpenAI to override that refusal.

But operational decisions belong to the government, not the technology provider. OpenAI can advise on technical capabilities and build safeguards. It cannot dictate policy.

The context matters. OpenAI’s Pentagon deal landed hours after rival Anthropic was blacklisted and designated a “supply-chain risk to national security” following failed negotiations with the Department of Defence.

Anthropic had been the first AI lab to deploy models across the DoD’s classified network and was negotiating ongoing contract terms when talks collapsed. The company sought assurances that its models wouldn’t be used for fully autonomous weapons or mass surveillance of Americans. The Pentagon wanted Anthropic to agree to deployment across all lawful use cases without such restrictions.

When Anthropic refused, Defence Secretary Pete Hegseth announced the company would be blacklisted, preventing any military contractor from conducting commercial activity with it.

OpenAI stepped into that vacuum the same day, announcing its own DoD arrangement. The timing drew fierce criticism, including from some OpenAI employees who viewed it as opportunistic.

STAY UPDATED: JOIN WHATSAPP CHANNEL

Altman has since acknowledged the optics were poor. In posts on X, he conceded the announcement “looked opportunistic and sloppy” and that the company “shouldn’t have rushed to get this out on Friday.”

Following the backlash, OpenAI amended its Pentagon agreement to add explicit language addressing surveillance concerns. The revised deal now states the technology shall not be intentionally used for deliberate tracking, surveillance or monitoring of US persons or nationals.

Altman also claimed the agreement prohibits use of OpenAI technology for mass domestic surveillance, autonomous weapons systems, and high-stakes automated decisions like social credit systems. He stated that intelligence agencies, including the NSA, will not use OpenAI’s tools.

However, the amended terms make no explicit mention of autonomous weapons, one of the key issues that fractured Anthropic’s relationship with the Pentagon. The language focuses on OpenAI technology not being used “to direct” such systems, leaving room for interpretation about supporting roles.

The Pentagon deal has triggered measurable commercial damage. A “QuitGPT” movement emerged immediately after the announcement, with users cancelling paid ChatGPT subscriptions in protest. Whilst the full scale remains difficult to quantify, the backlash appears to have influenced Altman’s decision to amend the contract terms and attempt public damage control.

WATCH MARKETING EDGE ONTV

OpenAI already held a $200 million Pentagon contract from last year, allowing the agency to use its models in non-classified applications. The new agreement extends that access to classified networks, placing OpenAI’s technology alongside sensitive national security operations.

The controversy highlights tensions facing major AI companies as their products move into high-stakes government and military applications. Technology providers want to maintain ethical boundaries whilst also pursuing lucrative government contracts. Governments want maximum flexibility in deploying capabilities they’ve purchased.

Altman’s Tuesday message to staff suggests OpenAI has chosen its position in that tension. The company will provide technology and technical guidance. It will build safety mechanisms and refuse certain tasks. But it won’t attempt to control how the military uses tools once they’re deployed.

Whether that stance satisfies OpenAI employees, many of whom entered AI development with concerns about weaponisation and surveillance, remains unclear. Altman has drawn a clear line: OpenAI builds the tools, the government decides how to use them, and employees who feel uncomfortable with that arrangement must either accept it or leave.

The legal battle between Altman and Elon Musk, who co-founded OpenAI before departing, is scheduled for trial next month. Musk’s rival AI company xAI did not respond to requests for comment on the Pentagon developments.

For OpenAI, the amended Pentagon contract represents an attempt to balance competing pressures: employee concerns about military applications, government demands for deployment flexibility, commercial consequences from user backlash, and competitive positioning against rivals like Anthropic who drew harder lines.

Altman’s all-hands message suggests which of those pressures won out. The government’s operational autonomy takes precedence. Everything else adjusts around that reality.