An OpenAI engineer has been arrested on the charge of planting a recording device in an executive conference room, concealing it behind a television in the apparent hopes of capturing private information from high-level confidential meetings.
This was not exactly a caper on the level of James Bond or Ocean’s 11, however. The recording device was plugged into a USB port behind the television and wrapped in toilet paper for some reason. The perpetrator also apparently used their spouse’s chatbot account to ask about how to build and conceal a recording device as well as entering prompts referencing the specific MAC address of the device they eventually planted.
Recording device planted in San Francisco office used for high-level meetings
The engineer targeted a conference room in OpenAI’s San Francisco office that was regularly used for high-level discussions of internal research, relationships with governments and business development, among other sensitive matters.
While the choice of placement of the recording device seemed to be savvy, nothing else about the scheme reflected the level of competence one would hope for from an OpenAI engineer. The accused, a Saudi national named Abdulrahman Alfozan who has been with the company for a little over two years, apparently decided to use his employer’s own chatbot products to plan the entire caper from the ground up.
This began with research into the basics of how to build and conceal a surreptitious audio recording device. The perpetrator apparently settled on a prefabricated USB device sold by an IoT hardware company called Seeed Studio. This research also included a week of intensive questioning of the chatbot as to the particulars and technical details of the operation, such as how to install and maintain access to the recording device and encrypting the contents of its attached SD card. Alfozan allegedly used the account of his spouse for these queries, but the Department of Justice press release notes that it shared “multiple identifiers and attributes” with his own personal chatbot account.
However, the recording device did seem to go undetected for about two days. The DoJ release says that it was planted on September 28 and not discovered by OpenAI security until September 30 during a bug sweep. The contents of the device’s SD card were encrypted, but the security team apparently identified Alfozan by planting a hidden camera behind the television and waiting for him to make an appearance. He was then captured on camera peering behind the television on October 5.
Alfozan also reportedly had a history of accessing and being around the conference room during this time despite not being authorized to be inside of it, as documented by security camera footage. Additionally, a security search of his personal bag on October 5 turned up a second recording device similar to the one that had been planted.
OpenAI says recording device did not capture anything sensitive
Though there was a window of potentially two days in which the recording device was operating without company knowledge, OpenAI says that it has “no indications” of “model weights, source code, or customer data” being accessed. It is unclear if the recording device was only saving locally to the SD card, or additionally transmitting in some way.
There is also no real indication of foreign government involvement as of yet. Though Alfozan is a Saudi national who obtained legal permanent US resident status in 2020, he studied at UC Berkeley and MIT dating back to the early 2010s and has since moved through a number of prominent roles at major US tech firms including Meta. A Fortune investigation notes some connection to a Saudi investment firm, one of which has a Saudi royal as a member, but there are no other known connections at this time. Alfozan and his family appear to have planned to flee the US after the October 5 investigation, however, booking one-way tickets to Saudi Arabia via Istanbul. Alfozan was arrested three days before they were scheduled to leave the US.
Alfozan has been charged with illegally intercepting oral communications, which carries a maximum penalty of five years in prison and a $250,000 fine. While indications thus far point to his case being one of attempted personal profit, there have been high-profile government-linked spy campaigns at tech companies in recent years. In 2024 a former Google software engineer was convicted of sharing the company’s trade secrets with Chinese companies, and another Saudi national was caught leaking private Twitter user data to the government of Saudi Arabia in 2022. Anthropic CEO Dario Amodei has issued a warning that frontier AI companies are likely infiltrated with foreign spies, particularly those working on behalf of the Chinese government.
John Bruggeman, vCISO of CBTS, notes that while this incident did not appear to lead to major impact it is nevertheless imperative that organizations not rely on attackers being “dumb”: “The takeaway for me is not ‘could our AI catch this’ or, ‘should I create an agent to query our calendar logs everyday?” The takeaway is, do we have the data to catch this kind of activity? Also, do you have anyone looking at the data or alerts before the physical security team, doing weekly sweeps, finds the bug? Internal queries by someone who isn’t an executive assistant about meeting rooms and executive calendars is the signal you would have to flag. You also have to hope your insider is dumb enough to use an internal tool to ask how to build a bug. The capability to correlate that doesn’t require a breakthrough model. It requires deciding, in advance, that this kind of cross-system behavioral correlation is worth building and staffing. You will need a human in the loop checking the alerts. If you’re a multi-billion dollar a year company, and you have valuable IP, it’s probably worth investing in a real insider risk program.”
