On Monday morning, chalk messages appeared on the sidewalk outside OpenAI’s San Francisco offices: “Where are your red lines?” “You need to speak louder.” “What are the guarantees?”
The writings, according to social media reports and media outlets, were put there by activists. However, many inside the building share some of these sentiments following OpenAI’s agreement on Friday with the Pentagon to utilize its AI models in classified systems.
Anthropic had already rejected a contract renewal with the Pentagon, feeling the wording didn’t align with the company’s mandates concerning the use of AI in mass surveillance and autonomous weaponry. The Pentagon subsequently blacklisted the company, labeling it a supply chain risk.
The contracts are replete with legal and technical intricacies. But on public forums and in closed conversations, OpenAI personnel are voicing feelings about how the OpenAI leadership navigated the Pentagon negotiations. Many employees “deeply respect” Anthropic for standing up to the Pentagon and are displeased with how OpenAI handled its own agreement, one current employee told CNN on condition of anonymity to speak freely.
As the clock ticked toward the Pentagon’s Friday deadline for Anthropic’s contract consent, OpenAI CEO Sam Altman surprised many by stating he agreed with his rival, Anthropic CEO Dario Amodei, and that he shared the same “red lines.”
Yet, it turned out Altman was simultaneously negotiating his own deal. Criticism erupted hours later when OpenAI announced its Pentagon contract, seemingly aiming to fill the vacancy left by Anthropic. After OpenAI published some contract terms on Saturday, many external observers immediately questioned how the “red lines” on autonomous weapons and mass surveillance would truly be upheld, with some arguing the phrasing still enabled those safeguards to be bypassed.
In reaction, Altman publicly addressed queries on X Saturday evening and announced on Monday that OpenAI had amended its Pentagon contract to more clearly define limitations prohibiting the use of OpenAI services in surveillance programs. (Autonomous weaponry was omitted from the added language he posted online.)
Many staff recognize the necessity of supporting the government as the U.S. competes with China in the AI realm, according to a current employee. But they also felt a contract of such significant scope and importance was rushed through.
“Part of it has to do with how it was perceived, how it was communicated, and what that narrative became,” the employee remarked.
Some personnel voiced their discontent publicly. Researcher Aidan Maclaughlin posted on X Monday morning before Altman’s contract update: “Personally, I don’t think this deal was worth it.” He later described the internal discussion on the topic as “overwhelming” but said he felt “incredibly proud to work somewhere where people can speak up.”
Jasmine Wang, who works on AI safety at OpenAI, wrote that she wanted an “independent outside counsel” to review the new contract text Altman announced Monday. She later reposted legal analyses that both supported OpenAI’s claim that it cemented their red lines and others that criticized it as “weasel words.”
Altman acknowledged a communication breakdown.
“The issues are extremely complex and require clear communication,” Altman posted on X Monday. “We truly tried to de-escalate and avoid a much worse outcome, but viewed through my lens, it just looked haphazard and clumsy.”
At an internal all-hands meeting at OpenAI on Tuesday, Altman reiterated that the hasty finalization of the deal was a “mistake,” according to a source familiar with the meeting. However, Altman stated that OpenAI cannot weigh in on individual use cases of its technology, such as differentiating which specific military operations might be deemed good or bad.
An OpenAI spokesperson directed CNN to Altman’s public statements.
Some employees on Tuesday were also frustrated that certain observers were framing Anthropic as a savior, despite years of working with the Pentagon and major defense contractor Palantir without similar scrutiny.
On Tuesday, Altman told staff that he believes governments should collaborate with labs like OpenAI that enforce safety standards, rather than companies with fewer safeguards in place. He mentioned that he is encouraging the government to drop Anthropic’s supply chain risk status.
“I believe we will have better models that encourage the government to be willing to work with us, even if our safety system irritates them, or imposes certain restrictions or whatever else,” he said at the meeting.
About the Author
