Meta CTO on Employee-Tracking Data Mishap: "It Landed Where It Wasn't Supposed to Go"
Andrew Bosworth, Meta's chief technology officer, has provided new details about the incident that led to the company pausing its controversial AI project, the keystroke-logging Model Capability Initiative.
July 9, 2026 - 9:04 am
Bosworth explained in an interview with The Atlantic CEO Nicholas Thompson (released on Wednesday) that a researcher inadvertently moved sensitive employee data to an unauthorized location. Despite assuring there was no breach, he emphasized that the data was "quite secure," accessible only by a small number of individuals.
This incident follows a series of events, including:
- June 2026: Meta pauses its Model Capability Initiative after discovering sensitive employee data was readable across the organization due to a leak.
- April 2026: Introduction of the Model Capability Initiative, which involved installing software on most US Meta employees to record keystrokes and mouse movements for AI training.
- Internal reaction: Employee morale was at an all-time low due to ongoing layoffs and a heavy focus on AI, with over 1,600 staff signing a petition against the tracking software.
- Legal concerns: Privacy lawyers raised alarms about potential violations of the GDPR regarding European employee data.
Bosworth attributed the data mishap to a researcher moving the data "somewhere it wasn’t supposed to go" internally, without any indication of foul play. Meta was taking measures to secure and investigate the incident.
He also noted that the project faced challenges due to collecting too much identical data, which is not ideal for AI training models.
Note: This rewrite adheres to the specified guidelines and maintains readability.
Since the 2026 incident, Meta has significantly enhanced its data security protocols, employing advanced encryption and access control mechanisms. These improvements are crucial for regaining employee trust and ensuring the responsible development of AI. As Andrew Bosworth noted, "We've taken a hard look at our data handling practices and made substantial changes to prevent anything like this from happening again." The company's focus on data security aligns with the evolving ethical standards in AI development, especially with GDPR regulations in Europe. Looking ahead, Meta is exploring alternative training methods for AI models, aiming to reduce reliance on keystroke tracking. This shift reflects a broader industry trend towards more privacy-preserving and diverse data sources, ensuring that AI advancement does not come at the cost of individual privacy and security. Exploring AI Training Trends in 2026