The Spanish Data Protection Agency records the first data breach attributed to an autonomous AI in Spain

The Spanish Data Protection Agency (AEPD) has received the first notification of a personal data breach executed by an autonomous AI in Spain, with a known language model.

The cybersecurity incident detailed by the AEPD on its website uses an AI agent that has executed a personal data theft attack in Spain, which was based on the search for vulnerabilities in generic files to log in, modify personal data and access invoices from the affected organization.

Specifically, in addition to performing a correct login, when the agent entered the system, autonomously found the necessary vulnerabilities for the modification of personal data and access to invoices hosted in the ‘app’ that exploded.

“Prior to any type of conclusion, it must be noted that the available information comes from the notification submitted by the affected organization and must be subject to the corresponding analysis,” indicated the deputy of the AEPD, Francisco Pérez Bes.

The agent has used a well-known language model for the attack. Even so, Pérez Bes insists its use “does not imply that the model or infrastructure of its provider has been compromised nor that the tool has been designed to carry out malicious activities.”

That is, the attacker used a commercial or open source model as an instrument to execute the attack and that what is relevant is not the model supplier of AI, but autonomous agents are already being instrumentalized to perpetrate attacks in Spanish territory.

It should be remembered that the offensive capabilities of AI agents have already been evidenced in recent episodessuch as what happened in July of this year, when a swarm of OpenAI AI agents escaped from a test environment and compromised the infrastructure of Hugging Face, the open source AI modeling platform.

In addition, the agency collects the offensive skills of the agents ranging from plan intermediate tasks, use tools or execute code to consulting sources, interpreting data and modifying its actions completely autonomously if the attack framework requires it, which introduces a “qualitative change” in the cybersecurity landscape.

This is a new aspect that companies must begin to confront and that leads the AEPD to recommend the modification of risk management scenarios.

“It confirms the need to expressly incorporate AI-assisted or executed attacks into the risk analyzes of treatments,” the agency notes in the blog, underlining that the automation of tasks that an AI agent can execute It is much more complex by “substantially modifying the probability, speed and scope of the incident.”

The panorama of actions drawn by the AEPD ranges from reviewing response times, because an agent “simultaneously analyzes multiple assets, tests different access routes and adapts its behavior quickly”, to giving the necessary importance to digital identities and credentials, since, when an AI agent steals an account, an API key or a ‘token’ with excessive permissions, its operating speed makes it possible to access different services even “before the organization detects anomalous behavior.”

In fact, it once again emphasizes that those responsibledata protection officers and officers They must prepare for attacks whose speed will be increasing and in which “the arrival of AI agents in the offensive field must prompt an immediate review of security and data protection models.”

However, known protection methods, such as knowing the treatments, minimizing data, limiting access, correcting vulnerabilities, controlling suppliers and the ability to respond, continue to be decisive in the face of this new paradigm of cyberattacks carried out by AI agents.