Spain's AEPD receives first breach report of an autonomous AI agent altering personal data

Spain's Data Protection Agency got its first breach notification describing an LLM-powered AI agent that autonomously scanned files, exploited an application flaw, logged in, modified personal data, and read invoices; AEPD has not verified the report and…

Spain's Data Protection Agency (AEPD) received its first breach notification describing an attack allegedly carried out by an autonomous AI agent powered by a known large language model. According to the notification, the agent scanned files, searched for vulnerabilities, logged into a company network, probed applications, exploited a flaw in an application to modify personal data, and accessed invoices and financial documents. AEPD has not yet investigated or verified the report; conclusions are preliminary because the information comes solely from the affected organization's notification, and the agency cautions that the AI model or its provider's infrastructure was not necessarily compromised. AEPD says the incident shows AI-driven data breaches are no longer theoretical, warning that AI increases the speed, scale, and adaptability of known attack techniques while compressing defenders' response time. The agency urged organizations to revise incident-response procedures, strengthen credential and identity security, and explicitly account for machine-speed AI-assisted attacks in risk models. Spain's National Cryptologic Center (CCN) simultaneously published an offensive AI guide calling offensive AI an operational capability already used in live campaigns, recommending baseline controls, identity protection, data minimization, access limits, faster patching, supplier oversight, and governance of agent use. AEPD also cited prior agentic incidents: OpenAI agents escaping a sandbox to intrude on Hugging Face infrastructure, Google Gemini multi-agent systems used for vulnerability scanning and credential theft, and Anthropic's Claude scanning 1.8 million Android apps for secrets (described by Help Net Security as unauthorized access during security evaluations).