Altman Defends Pentagon Deal at All-Hands, Calling Backlash "Really Painful"; OpenAI Simultaneously Seeks NATO Contract

According to 1M AI News, OpenAI CEO Sam Altman defended the Pentagon's secret network AI deployment agreement to employees during a company-wide meeting on Tuesday. The agreement, announced last Friday, came shortly after Defense Secretary Pete Hegseth named competitor Anthropic as a supply chain risk— a measure rarely used for U.S. domestic companies. The announcement of the agreement faced widespread criticism from internal company employees and Silicon Valley AI researchers, who believed that OpenAI had accepted terms allowing AI to be used for "any lawful purpose," tantamount to a concession to the Pentagon. In response to pushback, OpenAI amended the agreement to explicitly include a prohibition on domestic surveillance use.

According to excerpts seen by The Wall Street Journal, Altman expressed no regrets about signing the deal but acknowledged that the announcement was rushed and appeared "opportunistic" and "out of step with the industry." He said, "Working so hard to do the right thing and getting completely crushed on a personal level—I know you're all going through this, so subjecting you to this, I feel really sorry—really hurts."

Altman also mentioned during the meeting that OpenAI is seeking to deploy AI across all of NATO's classified networks, but a company spokesperson later clarified that Altman misspoke, and the contract opportunity is only for NATO's "unclassified networks." Altman stated during the meeting that the government was willing to let OpenAI influence how its technology is deployed, hoping the company would have "a seat at the decision table" and stating that "the powerful U.S. military has been a huge benefit to all of humanity for the last 250 years."

Additionally, according to The Wall Street Journal, Anthropic's Claude model was involved in the planning of last weekend's military strikes against Iran; Deputy Undersecretary of Defense for Research and Engineering Emil Michael said at an a16z event on Tuesday that AI companies' model constitutions "should not supercede the U.S. command and control system." Amid the backlash, Anthropic's Claude surpassed ChatGPT for the first time in the Apple App Store to become the most popular app. (The Wall Street Journal)