The Intercept reports that three senior OpenAI staff spoke with people posing as Ukrainian officials, discussing possible military AI support for Ukraine. OpenAI says it investigated the call, concluded it was a prank and acted to prevent similar approaches.
OpenAI Watch analysis
What happened
The Intercept says the 17-minute video call involved Sasha Baker, Caroline Zier and Ben Crampton, who work on national security policy and strategic initiatives. According to the publication’s account of a recording posted online, the conversation touched on offensive military uses of AI, contact with Ukraine’s Ministry of Defence and work with the Ukrainian government.
The publication quotes OpenAI spokesperson Nate Evans as saying the company had launched an investigation into the call’s authenticity and concluded it was a prank. Evans said OpenAI had acted internally and with relevant authorities to prevent similar situations. The account does not establish that any proposed service was agreed or provided.
Why it matters
The episode puts two difficult questions in the same room: how AI companies handle national-security approaches, and how easily a convincing impersonation can reach senior staff. The Intercept’s account suggests that people posing as officials got a substantive conversation. OpenAI’s response confirms its conclusion that the call was a prank, but does not address each reported discussion point.
Our read
This is a story about a reported security lapse and the boundaries of AI companies’ military work, not proof that OpenAI supplied offensive capabilities to Ukraine. The distinction matters; so does having a process that can verify who is on the other end before a sensitive conversation gets under way.
What to watch
- Whether OpenAI explains what safeguards it changed after the incident.
- Whether the company clarifies the reported Ukraine contacts and the scope of its military work.
- Whether further evidence or a response from the named staff changes the account.
Discussion spark: For sensitive government or military discussions, should AI companies require formal identity checks before any substantive meeting, even if that makes legitimate access slower?
Sources and evidence
- OpenAI Fooled by Russian Pranksters Into Thinking They Were Speaking to Ukrainian Official – The Intercept (11 November 2026, 23:53 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.