The AEPD has published a notification in which all the phases of the intrusion — connection, vulnerability search, data modification, access to invoices — have been chained together by an agent.
On September 16, 2026, the Spanish Data Protection Agency published the details of a breach notification that is out of the ordinary: according to the victim organization, the entire attack was carried out by a software agent, without human intervention step by step. The agent first logged into the system, looked for weaknesses in the application on its own, exploited one, then modified personal data and viewed invoices.
Until now, AI has appeared in incidents as an accelerator: writing phishing emails, deepfakes, industrializing campaigns already designed by humans. What changes here is attribution: the AEPD speaks of a third party having used an agent as an instrument to chain the different phases, and qualifies the episode as a qualitative change. The agency has not named either the model used or the organization affected, and has not yet verified the elements transmitted itself.
The technical hypotheses remain open: circumvention of the safeguards of a business model, escape from a test environment, or tailor-made assembly around a large language model. For compliance officers, the practical consequence lies elsewhere: risk analysis grids and incident response times have been calibrated on human attackers, whose pace is no longer the reference.
Read the full article on the original source
Go to the source →