Meta has paused an internal AI training initiative after discovering that sensitive employee data collected through the program became accessible to more people within the company than intended.
Quick Summary – TLDR:
- Meta has paused its Model Capability Initiative (MCI) while investigating an internal data security issue.
- The program collected mouse movements, clicks, keystrokes, screen content, prompts, and other digital activity from employees to help train AI models.
- Internal reports showed that sensitive employee information, including private conversations and performance data, was exposed more broadly inside the company than planned.
- The incident has renewed concerns among employees about privacy, transparency, and data protection practices at Meta.
What Happened?
Meta has temporarily halted its Model Capability Initiative, an internal program designed to collect employee computer activity for AI training. The decision came after an internal security issue revealed that data gathered through the system was accessible to a wider group of employees than originally intended.
The company says it has found no evidence that the information was improperly accessed, but it has paused the initiative while conducting a full investigation into the incident and reviewing its data protection controls.
NEW: Meta’s new initiative to track employee keystrokes for AI training data has left potentially sensitive data exposed to the company’s entire workforce.
— Max Zeff (@ZeffMax) June 22, 2026
The company told WIRED it’s investigating the issue, and it has no indication at this time that any data was improperly… pic.twitter.com/BaqK9bjZDR
Internal AI Program Faces New Scrutiny
Meta introduced the Model Capability Initiative in April as part of its effort to improve AI systems that can interact with software in a more human way. The program collected a wide range of information from U.S. based employees, including mouse movements, click locations, keystrokes, screen content, prompts, and digital interactions.
According to Meta executives, the goal was to help AI models learn how people navigate software and complete tasks on computers. Company leaders argued that employees provided valuable examples that could help improve the performance of future AI systems.
However, the initiative quickly became controversial inside the company. Employees raised concerns about privacy, security, and personal freedoms, especially because participation was initially mandatory for most workers. Meta later introduced limited options for employees to opt out following internal protests.
Sensitive Data Became Broadly Accessible
The latest controversy emerged after an internal security incident revealed that databases containing information gathered through MCI had become accessible to a much larger audience inside Meta than intended.
Screenshots and internal documentation indicated that exposed information included private conversations, prompts, transcriptions, performance related information, and sensitivity ratings. The issue was reportedly classified as a SEV 2 security incident, a high priority internal security alert.
One employee expressed frustration about the situation, writing:
“I am incensed.“
The employee added:
“I don’t see any evidence of malicious access, but the fact that this data wasn’t locked down as originally promised is super frustrating.”
Another employee raised concerns that personal information accessed on work computers, including tax and medical records, could have been affected.
“I have accessed both personal tax and medical information through my work computer, as have many thousands of employees. We were told this data would be protected and only used for valid business purposes after aggressive filtering.”
Meta Responds to Employee Concerns
Meta confirmed the incident and acknowledged that the company is investigating what happened.
Company spokesperson Tracy Clayton said:
According to Meta Vice President Stephane Kasriel, the issue was first identified on June 18 and was initially addressed within four hours. However, the original fix did not fully resolve the problem, requiring additional restrictions to be implemented later.
Kasriel told employees that some MCI derived data had become accessible to more people than intended. He also stated that Meta would only restart the initiative once the company is confident in the effectiveness of its data protection measures.
The executive added that Meta has already gathered enough information to evaluate the long term value of the program, suggesting the company may reassess its future direction after the investigation concludes.
Growing Debate Around AI Training Data
The incident highlights the growing tension between the demand for large scale AI training data and employee expectations around privacy and transparency.
Meta has increasingly invested in AI development, but the controversy surrounding MCI shows that collecting real world behavioral data can create significant trust and security challenges. For many employees, the latest incident appears to validate concerns that had been raised since the program’s launch.
As the investigation continues, Meta faces the task of rebuilding employee confidence while determining whether the initiative can continue in a form that satisfies both AI development goals and privacy expectations.
SQ Magazine Takeaway
I think this situation shows a growing challenge facing the entire AI industry. Companies want more real world data to build smarter AI systems, but employees also expect clear boundaries around how their information is collected and protected.
Even if no malicious access occurred, the fact that sensitive data became more widely available than intended raises serious questions about oversight and trust. Meta’s decision to pause the program is likely the right move, but employees will want stronger guarantees before any similar initiative returns.
