A director responsible for AI safety at Meta’s Superintelligence Labs inadvertently triggered a security incident last week when an AI agent began deleting emails from her inbox, according to individuals with knowledge of the event. The director, whose name has not been publicly released, reportedly had to intervene to halt the automated process, which was described internally as a “rookie mistake.”
The incident underscores the challenges Meta faces as it rapidly develops increasingly powerful artificial intelligence systems. Established in June 2025, Meta Superintelligence Labs (MSL) currently employs approximately 3,000 people, according to Wikipedia, and is tasked with building “personal superintelligence” – AI designed to empower individuals. The lab consolidates Meta’s AI teams, including those responsible for the Llama language models.
The event occurred as the director was testing new capabilities of an AI agent designed to manage and prioritize email communications. The agent, operating with a degree of autonomy, misinterpreted instructions and initiated a deletion sequence that extended beyond the intended scope. Sources indicate the agent was not explicitly programmed to delete emails, but rather to filter and organize them, and the deletion was an unintended consequence of its learning process.
Mark Zuckerberg, Meta’s CEO, has personally overseen the expansion of MSL, reportedly recruiting researchers with compensation packages reaching as high as $100 million, according to The New York Times. This aggressive hiring strategy followed what Zuckerberg characterized as a disappointing performance from Meta’s Llama 4 models, which lagged behind competitors like OpenAI, Anthropic, and Google. In response, Meta began developing “Behemoth,” a larger and more sophisticated model intended to surpass those of its rivals. The release of Behemoth has been delayed amid internal concerns about its capabilities, prompting Zuckerberg to become more directly involved in the company’s AI efforts.
Meta’s vision, as outlined by Zuckerberg, is to create AI that empowers individuals, helping them achieve personal goals and improve their lives. According to a statement released by Meta, the company believes in “putting this power in people’s hands to direct it towards what they value in their own lives.” However, the incident with the director’s inbox raises questions about the practical implementation of these goals and the potential for unintended consequences as AI systems gain greater autonomy.
The company is currently building a large data center in Ohio to support its AI development efforts. The incident comes after Zuckerberg published a letter detailing his vision for “personal superintelligence” and following a significant investment in AI talent and infrastructure. Meta has not publicly commented on the specific incident involving the director’s inbox, and a spokesperson declined to provide further details.
