Russian Pranksters Fooled OpenAI's National Security Team With a Fake Ukrainian Official Call

A Russian comedy duo reportedly got three of OpenAI's senior national security staff to talk candidly about arming Ukraine, by pretending to be a Ukrainian intelligence chief. The Intercept reported the story on November 11, 2026. OpenAI confirms the call was a prank and says it investigated "weeks ago". The episode is small in dollar terms, but it shows how a company that sells AI to governments can be reached through plain social engineering.

This post explains what is confirmed, what rests on a recording we cannot verify ourselves, and what any team handling sensitive partner talks should copy. For related background, see our coverage of OpenAI's takedowns of Russian influence operations and of OpenAI's national security hire Dan Driscoll.

TL;DR

What did The Intercept report?

According to The Intercept, Vovan and Lexus arranged a high-level meeting with OpenAI by pretending to be Rustem Umerov. The duo uploaded a recording of the 17-minute video conversation to Rumble. The three OpenAI participants were Sasha Baker, who runs national security policy, Caroline Zier, who leads strategic initiatives, and Ben Crampton, who leads European national security work. The report says Baker and Zier previously held senior Department of Defense roles, and Crampton worked in the European Union and United Nations.

The Intercept's account of the conversation covers several topics:

• Military support for Ukraine. The participants discussed OpenAI's support for Ukraine's defense, including the possibility of offensive capabilities.
• Staff attitudes. Crampton is quoted saying there is "a general reluctance on the part of many people who are senior in the company" to get involved in offensive military uses of OpenAI's technology.
• Existing contacts. Baker reportedly said the company has been in contact with the Pentagon about helping Ukraine. Crampton described contact with Ukraine's Ministry of Defense through a Washington colleague.
• Daybreak Blue. Officials mentioned work with the Ukrainian government around the Daybreak Blue model, a system for fixing computer vulnerabilities. Our explainer on the Daybreak Red and Blue tiers covers what that product is, and OpenAI describes the program on its Daybreak page.
• Politics. Baker is quoted saying OpenAI cannot help with Elon Musk, who "is not a fan of our CEO either", and reassuring the caller that most of the U.S. government and public support Ukraine.

Treat these as the reporter's account of a recording made by people with a political motive. The Intercept does not say it independently authenticated the video, and explainx.ai has not heard it. OpenAI's own statement confirms the call itself was real and was a prank.

How did two comedians get a meeting with OpenAI?

The method is the most instructive part. Vladimir Kuznetsov, the "Vovan" of the duo, told The Intercept that it was simple. They first contacted a senior Ukrainian official in San Francisco while posing as Umerov. That official, it turned out, was already trying to build links with OpenAI through the Ukrainian diaspora in California, and was already in touch with Baker. Believing he was speaking with the real Umerov, the official offered to arrange a meeting with OpenAI.

So the chain was:

1. Impersonate a famous official to a real mid-level contact.
2. Let that contact vouch for the "official" inside the target company.
3. Reach the target through a warm introduction instead of a cold approach.

This is a transitive-trust attack. OpenAI's staff did not verify a stranger. They trusted a known intermediary, who had trusted the pranksters. Both Baker and Zier reportedly said they had met Umerov before and were glad to speak to him again, yet the fake Umerov kept the camera off for the entire call. That is a red flag in hindsight, and a standard one: a person claiming a prior acquaintance who will not show a face.

Notably, the report does not describe AI voice cloning or face swapping. We flag this because it is the first assumption many readers make.

Who are Vovan and Lexus?

Vladimir Kuznetsov and Alexei Stolyarov, known together as Vovan and Lexus, are Russian comedians noted for prank telephone calls to prominent people. According to The Intercept, they have been doing this since 2011, and past victims range from Elton John to John McCain. Their pranks have often aligned with Russian foreign policy aims. The report says they have been accused of working with the Russian government, which they deny.

Whether they are comedians, influence operators, or both, the effect for the target is the same: a recorded private conversation, edited and published at a moment of the pranksters' choosing. That matters for a company in OpenAI's position, because the content of the call touches three sensitive areas at once: war, defense contracts and a feud with a rival CEO.

What did OpenAI say?

OpenAI spokesperson Nate Evans gave The Intercept this statement: "Weeks ago, we launched a thorough investigation into the call's authenticity. We acted internally and with the relevant authorities to conclude that it was a prank conducted by individuals with a history of targeting officials and other companies. We took action to prevent similar situations from happening in the future."

Three things follow from the wording. First, OpenAI knew about the call before the recording was published, so the timeline is weeks, not hours. Second, it says it involved "relevant authorities", but does not name them. Third, "took action" is not defined: we do not know whether that means new verification rules, a training session, or changes to who takes such meetings. The Ukrainian Embassy in Washington did not immediately respond to The Intercept's request for comment.

Why this matters beyond the embarrassment

Shield with a star representing OpenAI national security and military AI policy under pressure

OpenAI has moved deep into government work. The Intercept notes an ongoing 200 million dollar contract with the U.S. Department of War, which according to contract documents obtained by the outlet covers AI applications for "warfighters". We covered the wider tension between labs and the Pentagon in our piece on Emil Michael and lab cooperation. In that setting, the people who talk to foreign governments are security-relevant by role. A private call that reveals internal attitudes, such as senior staff being reluctant about offensive uses, is useful to adversaries and to critics alike.

Four points worth separating:

• Not a model leak. Nothing in the report says weights, code or customer data left OpenAI. The exposure is conversational: internal views and the existence of contacts.
• Not a policy statement. Staff remarks on an unverified call do not change OpenAI's published position on military use. Check OpenAI's own usage policy and statements for that.
• Reputational and political risk. The call touches Ukraine, the Pentagon, President Trump's stance on Zelenskyy and Musk. Edited clips can travel fast.
• A reminder about targeting. The same Russian-linked ecosystem that pranks officials also runs influence campaigns that use AI tools, which OpenAI has described in its own threat reports.

What should teams copy from this?

Any AI company that talks to governments, or any team that holds sensitive partner calls, can use a short checklist. These are explainx.ai's suggestions, not a standard from OpenAI or The Intercept.

If you run AI agents that join meetings, take calls or send messages on your behalf, the same attack applies to them with less friction, because an agent can be talked into acting on a convincing identity claim. AgentBeam, the agent security platform from the explainx.ai team, is built to stop AI agents before they take dangerous actions, which is one control layer for that risk.

How does this compare with AI-driven impersonation?

The OpenAI call looks low-tech, and that is the point. Many impersonation attacks do not need cloned voices. Still, the AI-enabled version is growing. Our reports on AI synthetic voices in 911 swatting, a DOJ indictment over AI impersonation of a CEO and lawyer and a Russian-linked AI deepfake video campaign show how voice and video fakes lower the cost of this trick. A defense that works against a camera-off human prankster, such as an independent callback, also works against a cloned voice.

What is confirmed and what is not

Confirmed by OpenAI's statement to The Intercept: a call took place, it was a prank by people with a history of targeting officials and companies, and OpenAI investigated and acted.

Reported by The Intercept from the pranksters' recording and a Kuznetsov interview: the names of participants, the Umerov impersonation, the 17-minute length, the topics, the quotes and the route through a Ukrainian official in San Francisco.

Not established: whether the published video was edited, which authorities OpenAI contacted, what "action" it took, whether any contract with Ukraine exists, and whether the real Umerov or the Ukrainian government has a view. No Ukrainian reply had been published when the report appeared.

Bottom line

A 17-minute call with a face-off "official" reached the top of OpenAI's national security staff because a human intermediary vouched for the caller. The fix is process, not model safety: verify out of band, treat warm intros as unverified, and assume every sensitive call is recorded. We will update this post if OpenAI or Ukraine's government publishes more.

Related reading

• OpenAI disrupts Russia and Iran false-front influence operations
• OpenAI hires Dan Driscoll as national security counselor
• GPT-5.6-Cyber and Daybreak Red and Blue tiers
• Emil Michael on lab cooperation with the Pentagon
• AI synthetic voices in 911 swatting
• DOJ indicts three over AI impersonation fraud
• Storm-1516 deepfake video and Russian influence

Details in this post reflect reporting available on November 12, 2026, and may change as OpenAI, Ukraine or The Intercept publish more.