OpenAI claims its AI busted a spying employee, but that might not be true

OpenAI claims it used its internal AI models to investigate and link a recording device planted in an executive conference room to one of its employees. But this claim is likely far-fetched.

An incident involving one of OpenAI’s former engineers, Abdulrahman Alfozan, who allegedly spied on company executives, has drawn significant attention, particularly after the AI company made a bold claim.

While OpenAI’s internal models are most probably sophisticated and powerful, it's unlikely that ChatGPT could do a better job of cracking this case than the FBI could.

While there could be a shred of truth in OpenAI’s claims, cybersecurity advisor Adam Pilton suggests we take it with a pinch of salt.

“The key investigative work was done by a human,” and the company’s models may have taken on a supporting role,” Pilton told Cybernews.

Positioning the company’s AI as the driving force behind the investigation could help OpenAI turn the incident in its favor.

“From what we know of the frontier AI companies, they never miss an opportunity for marketing, and for some, this incident may be viewed as a marketing gift.”

OpenAI can’t check for fingerprints, but it could destroy them

Instead of pioneering the investigation, OpenAI could have hindered it by mishandling the USB recording device and destroying physical evidence, such as DNA and fingerprints, Pilton told Cybernews.

Pilton also echoes the idea that the real evidence may have been derived from the accused employees' chatbot account.

“In terms of the device, it’s very possible that the accused employee vibecoded it, which, if so, would also provide evidence for OpenAI. Depending on the employee’s storage settings, these vibecoding chats will be stored and easily accessible to OpenAI,” Pilton concludes.

Furthermore, a start-to-finish investigation conducted by AI would likely not hold up in court.

How OpenAI’s models could’ve helped

The company claims that its internal models successfully identified the device and traced its origins.

When law enforcement investigates a digital device at a forensic level, they seek evidence linking an alleged individual to the crime. This evidence could be audio, visual, search, or prompt history.

While an AI model could analyze certain elements, like lines of code, coding style, and development history logs, authorities couldn’t arrest a criminal based on this information alone.

However, what would be more insightful is OpenAI’s access to its users' chat information.

Alfozan allegedly used a “chatbot” to vibecode spytech

Alfozan didn’t work alone. Importantly, the complaint alleges that a “chatbot account” associated with Alfozan had asked how to create a device similar to the one found in the OpenAI conference room.

Between September 24th and September 30th, Alfozan allegedly asked “a chatbot” how to not only build, customize, and install an audio recording device, but also how to remotely access the device and encrypt its recordings.

The commands included terms like “build a continuous audio recorder,” “customize WiFi . . . connectivity,” “improve audio and live listening,” and “encrypt the SD card.”

While the chatbot used to apparently vibecode the device used to spy on OpenAI executives hasn’t been named, OpenAI employees are all given free access to the company’s suite of AI models.

Cybernews has reached out to OpenAI and Sherwin Wu, the company’s head of technical staff, for comment.