Ethics flag at Pentagon AI office
Reporting says a defence official who oversees AI work sold millions in xAI stock after the Pentagon entered an agreement with the company, creating an apparent conflict-of-interest problem. The episode has amplified worries that governance and procurement integrity are becoming part of how contractors prove they can deliver ‘responsible AI’ to the DoD (The Guardian).
A Pentagon official who helps steer military artificial intelligence policy sold stock in Elon Musk’s xAI for a gain reported at between $5 million and $25 million after the Defense Department entered agreements with the company, according to reporting published on April 9. The official is Emil Michael, the under secretary of defense for research and engineering, a job that includes oversight of key artificial intelligence work. (theguardian.com) The timing is what set off alarms. The Pentagon’s Chief Digital and Artificial Intelligence Office announced on July 14, 2025 that xAI was one of four companies awarded separate agreements with ceilings of $200 million each to bring “frontier” artificial intelligence tools into Defense Department work. (ai.mil) Those four companies were xAI, OpenAI, Google, and Anthropic. The Pentagon said the deals were meant to speed up military use of large language models, automated software agents, and related cloud systems across warfighting, intelligence, and business operations. (ai.mil) (defensescoop.com) Michael was not a random bystander inside the building. Jacobin reported on March 26, 2026 that he had become a central figure in Pentagon decisions affecting artificial intelligence vendors, including a dispute involving Anthropic, while holding investments in rival technology companies. (jacobin.com) Federal ethics law is built around a simple rule: a government employee is barred from participating personally and substantially in a matter that has a direct and predictable effect on the employee’s financial interests. That criminal rule sits in 18 United States Code section 208. (law.cornell.edu) (oge.gov) There is a second rule about appearances, and that is where this story gets harder to shrug off. Federal ethics regulations say employees must avoid situations where a reasonable person would question their impartiality, even in cases that do not fit the criminal conflict statute exactly. (ecfr.gov) (law.cornell.edu) One detail in the reporting matters a lot: Michael reportedly received a certificate of divestiture before selling. That certificate is an Office of Government Ethics mechanism that lets an executive branch employee defer capital gains tax when selling an asset to comply with conflict-of-interest requirements. (theguardian.com) (ecfr.gov) (oge.gov) That does not erase the underlying question. A certificate of divestiture is a cleanup tool for disposing of a conflicted asset, not a public finding that the official’s earlier involvement was harmless or that no appearance problem existed. (ecfr.gov) (oge.gov) The xAI deal was already controversial before this stock-sale story landed. In September 2025, Senator Elizabeth Warren called the Pentagon’s xAI arrangement “uniquely troubling” and pressed Defense Secretary Pete Hegseth over procurement, accountability, and the use of Grok inside military systems. (defensescoop.com) That is why this is turning into more than one official’s disclosure problem. When the Defense Department buys artificial intelligence, it is not only buying code and computing power anymore; it is also testing whether the people awarding the work can show clean procurement, credible recusals, and ethics controls strong enough to survive scrutiny. (ai.mil) (ecfr.gov)
