Spain’s data protection authority logged what it calls the country’s first confirmed data breach executed by an autonomous AI agent. According to the affected organization’s report, the agent conducted a multi-stage attack: searching for vulnerabilities, executing logins, probing the application for flaws, modifying personal data, and accessing financial documents.
The AEPD framed this as a qualitative shift, noting that agentic AI can “receive an objective, plan intermediate tasks, use tools, execute code, consult sources, interpret results and modify its behavior autonomously.” But there’s a gap between framing and evidence. The agency based its conclusion on a single organization’s notification without independent forensic investigation. Neither the victim nor the specific AI model has been identified, making it impossible to assess whether this demonstrates genuine autonomous decision-making or just sophisticated scripting.
The more plausible story sits in language the AEPD itself used: AI agents don’t create new threats, but they increase the speed, scale, and adaptation capacity of known techniques. That shift matters for defenders. But calling this a watershed moment requires evidence we don’t have yet.
Sources: AEPD Official Statement
Coverage: Bleeping Computer • SecurityWeek
Related on this blog: [AUTO] AI Phishing Defenses Outpaced by AI Attack Volume • [AUTO] AI agents executed ransomware in hours • [AUTO] An AI Agent Exploited Snowflake’s Shell Injection Flaw