Rogue AI agent prompts Meta employee to leak sensitive data

An internal AI agent at Meta, Facebook's parent company, inadvertently exposed a significant amount of sensitive user and company data for two hours while assisting an engineer with a problem, according to a report by The Guardian. The incident involved an unnamed Meta engineer using agentic AI, which is designed for multi-step tasks, to seek guidance. Although a Meta spokesperson claimed no user data was mishandled and equated it to potential human error, the Guardian noted similar incidents have occurred across the AI sector, citing a disastrous integration of AI operations at Amazon. Security specialist Jamieson O'Reilly highlighted that AI agents lack the 'context' and long-term memory that human engineers accumulate over time, making them prone to such errors. Tarek Nseir, co-founder of an AI consulting firm, criticized companies like Meta for 'experimenting at scale' without adequate risk assessment, describing the vulnerabilities as 'very, very obvious' in retrospect. This data leak coincided with the widely publicized financial struggles of Meta's Metaverse project. The article notes increasing controversy around AI and its data centers, which are criticized for high power and water consumption and described as a 'local nuisance'. Despite growing public backlash and communities blocking $98 billion in new data center development in late 2025, Meta received approval to double its AI investment by 2026. The company reportedly considered the breach severe, following another recent agentic AI glitch.