Russian pranksters trick OpenAI leaders into discussing military AI

The prank call details

OpenAI leaders recently fell victim to a sophisticated prank involving a video call with two Russian comedians. The pranksters, who use the monikers Vovan and Lexus, successfully convinced three of the company's national security and strategic initiative leaders that they were high-ranking Ukrainian government officials. The interaction lasted for 17 minutes before the deception was revealed.

According to the Intercept, the three OpenAI staffers involved were Sasha Baker, Caroline Zier, and Ben Crampton. During this time, the company leaders spoke openly about the potential for military AI applications. They specifically referenced ongoing discussions with the US Department of Defense regarding the provision of AI tools to Ukraine.

The comedians recorded the entire meeting and have since made the footage available to the public. This incident has caused significant embarrassment for the AI lab, particularly as it seeks to establish itself as a major player in the defense sector. The pranksters have a history of targeting public figures, including notable politicians like John McCain and Bernie Sanders.

Sensitive internal discussions

During the call, Ben Crampton made several specific comments regarding the company's stance on offensive military technology. He noted that while the company wants a deep relationship with Ukraine, it might not be possible to reach their desired goals immediately. He also described the role of national security staffers as smoothing over executive reservations about military AI.

Crampton explained that many leaders at OpenAI would not want their technology used in offensive military operations. He stated that the task of the security team is to explain that the line between offense and defense can be conditional in certain situations. These comments provide a rare glimpse into the internal debates regarding the ethics and applications of the company's technology.

Furthermore, Crampton admitted during the call that OpenAI maintains a contact within the Ukrainian Ministry of Defense. He explained that this connection is primarily through a colleague based in Washington who previously worked with the Delta team in that ministry. This level of detail regarding specific government contacts was shared with the pranksters under the guise of official business.

How the breach occurred

The pranksters revealed that arranging the meeting was remarkably simple. They did not need to hack into any systems or use complex technical exploits to reach the OpenAI leadership. Instead, they utilized a social engineering approach by contacting a senior Ukrainian official who was currently working in San Francisco.

This official reportedly offered to facilitate the meeting between the comedians and the OpenAI team. By using a legitimate intermediary, the pranksters were able to bypass standard security filters that might have flagged a direct cold call from a Russian entity. This highlights a vulnerability where trusted third parties can be manipulated to provide access to high-value targets.

The incident is particularly concerning because OpenAI maintains a massive contract with the Pentagon. This contract involves deploying advanced AI systems in classified environments. If pranksters can successfully trick leaders into a meeting for a joke, it suggests that malicious actors could potentially achieve similar results to gather intelligence or influence policy.

OpenAI's official response

Following the incident, OpenAI acknowledged the breach and launched a thorough investigation into the authenticity of the call. The company stated that they acted internally and worked with the relevant authorities to conclude that the meeting was a prank. They noted that the individuals involved have a history of targeting officials and other corporations.

The company confirmed that they have taken specific actions to prevent similar situations from happening in the future. While the company did not release specific details about the new security protocols, the incident serves as a public reminder of the risks associated with video conferencing. It underscores the need for rigorous identity verification when discussing sensitive government contracts.

The fallout from this prank puts OpenAI in a difficult position as it seeks to navigate the complexities of military AI. The company must balance its desire to work with the Department of Defense with the need to protect its internal communications. This event may lead to stricter gatekeeping for all future interactions with foreign government entities.

Implications for AI security

This story highlights a growing concern regarding the security of high-level corporate communications in the AI industry. As companies like OpenAI become central to national security infrastructure, they become primary targets for both pranksters and state-sponsored actors. The ease with which the comedians gained access suggests that social engineering remains a potent threat.

For the broader tech industry, this serves as a warning about the risks of