OpenAI Claims to Have Discovered a Hidden Microphone, FBI Seizes It

OpenAI claims to have discovered a hidden recording device in one of its offices and says it has identified the perpetrator using its own models. The matter has been referred to the FBI, which is investigating, while the company asserts that it has found no unauthorized access to its technical assets or customer data. This sequence of events comes after separate and disputed layoffs involving three former employees.

Recent Disputed Layoffs Presented as Separate

OpenAI places this announcement in a context where recent departures were, according to the company, unrelated to the incident involving the device. The company had previously laid off three employees from its security and alignment teams, whom it accused of violating clear policies regarding the handling of sensitive information. The individuals involved, Jasmine Wang, Mikita Balesni, and Tomek Korbak, published a letter contesting certain allegations and expressing concerns about the lack of clarity surrounding their departures and its potential effects on internal transparency. OpenAI has maintained its position, via an internal note stating that an investigation uncovered a significant breach of trust, beyond what their letter describes, and that it supports the decision not to continue their employment.

Discovery of a Device and Referral to the FBI with Ongoing Investigation

OpenAI claims to have found a sophisticated recording device hidden in one of its premises and says it has referred the matter to the FBI. The company expresses gratitude to agents and prosecutors for a swift arrest and adds that the federal investigation is ongoing, to which it intends to continue providing assistance. OpenAI states that it has terminated the identified employee and communicated about the matter in a message posted on X on a Tuesday.

Identification Attributed to OpenAI Models, Without Technical Details

The company states that it used its own models to understand the device, trace its origin, and identify that an employee had installed it a few days prior. OpenAI claims to have found no evidence of access or manipulation of model weights, source code, or customer data. The company did not specify how these models were utilized to achieve the identification, and a spokesperson did not immediately respond to a request for comment. Nevertheless, OpenAI presents this case as an example of AI's ability to respond to sophisticated security threats.