OpenAI is amending its hastily struck deal to supply artificial intelligence to the US Department of War (DoW) after the chief executive of ChatGPT owner admitted it looked “opportunistic and sloppy”.
The contract raised fears that the San Francisco startup’s AI could be used for domestic mass surveillance, but its boss Sam Altman said Monday night that the startup would explicitly block its technology from being used for that purpose or deployed by Defense Department intelligence agencies like the National Security Agency (NSA).
OpenAI, whose ChatGPT has more than 900 million users, struck the deal almost immediately after it ousted the Pentagon’s existing AI contractor, Anthropic.
Anthropic had asserted that “using these systems for large-scale domestic surveillance is inconsistent with democratic values”, leading US President Donald Trump to call Anthropic “leftwing nut jobs” and instruct the federal government to stop using their technology.
Despite OpenAI’s denial that the agreement permitted surveillance use, commentators anticipated the Snowden scandal that broke in 2013, when it emerged that the NSA had engaged in mass harvesting of phone and Internet communications.
The deal prompted an online backlash against OpenAI, with users on X and Reddit encouraging a “Delete ChatGPT” campaign. a post Reading: “You are now training a war machine. Let’s see proof of cancellation.”
Chatbot Cloud, created by Anthropic, surged to the top of Apple’s App Store charts, toppling ChatGPIT, according to analysis by Sensor Tower.
In a message to employees reposted on X, the OpenAI CEO said the original deal announced Friday was done too quickly after Anthropic was ousted.
“We should not have rushed to publish it on Friday,” Altman wrote. “The issues are highly complex, and demand clear communication. We were really trying to minimize things and avoid a very bad outcome, but I think this just seemed opportunistic and dirty.”
When announcing the deal, OpenAI said the contract “has more guardrails than any previous agreement for classified AI deployment, including Anthropic”.
However, the US military’s use of AI has worried OpenAI and about 900 employees of Google, also a leading force in the technology, which have signed an agreement. open letter Called on their masters to forbid the DoW from using their products for surveillance and autonomous killing.
Warning that the US government is “trying to divide each company out of fear that the other will cave in”, he wrote: “We hope our leaders will put aside their differences and stand together to reject the DoW’s current demands for permission to use our model for domestic mass surveillance and to kill people autonomously without human oversight.”
The letter is signed by 796 Google employees and 98 OpenAI employees. OpenAI said in a blog post The DoW announced the deal, saying there is “no use of OpenAI technology to guide autonomous weapons systems” if its red lines are met.
However, observers, including Miles Brundage, OpenAI’s former head of policy research, have questioned how OpenAI has managed to secure a deal that overcomes ethical concerns that Anthropic believed were insurmountable. Posting on
Brundage said: “To be clear, the OAI is a complex organization, and I think many of the people involved have worked hard for a fair outcome. Some others I don’t trust at all, especially as it relates to dealings with government and politics.”
He also wrote in his ex-post that he would “prefer to go to jail” rather than obey the government’s unconstitutional order.
“We want to work through democratic processes,” Brundage wrote. “The government must make important decisions about society. We want a voice and a seat at the table where we can share our expertise, and fight for the principles of freedom.”
Meanwhile, three more US cabinet-level agencies – State, Treasury and the Department of Health and Human Services – have moved to stop using Anthropic’s AI products after the DoW declared the company a supply chain risk. Following the decision by Defense Secretary Pete Hegseth, Trump has ordered all US government agencies to phase out the use of Anthropic.
