Reuters informs that Spain’s data protection authority, AEPD, has received what it describes as its first personal data breach notification involving an AI agent apparently acting autonomously during a cyberattack.
According to the preliminary investigation, the attacker gained access using valid credentials. From there, an AI agent appears to have performed several stages of the attack: explored the system for vulnerabilities, modified personal data, and accessed accounts. Who initiated the attack and how the credentials were obtained remain unknown.
What changes when attackers become agentic?
The case is still under investigation, so conclusions about the precise role and degree of autonomy of the AI agent would be premature.
But the incident highlights a security challenge that is becoming difficult to ignore: AI agents can compress multiple stages of an attack into a faster, more scalable process.
“What AI agents change is the speed and economics of performing cyberattacks. Tasks that previously required significant time and coordination can therefore be executed faster and at a greater scale, allowing a relatively small number of attackers to probe more systems and pursue more attack paths.”
– Henry Patishman, Executive Vice President of Identity Verification Solutions
Authentication is a checkpoint, not a permanent trust decision
The fact that the reported activity began with valid credentials raises a broader question about what happens after authentication.
A successful login establishes that the required authentication evidence was presented. But it does not guarantee that every action performed during the session is legitimate.
Once a verified account or credential has been compromised, malicious activity may initially appear to be from a user the organization has already decided to trust. AI agents can make that problem harder by moving quickly after access is obtained: exploring permissions, identifying vulnerabilities, and interacting with multiple parts of a system before traditional controls recognize a pattern.
“A strong credential is only one part of a trusted transaction. The level of assurance ultimately depends on the entire chain around it: how the credential is presented and processed, how authentication is communicated, how the session is established, and what access is granted as a result. If attackers find a weakness elsewhere in that chain, they may not need to compromise the credential itself. As credentials become stronger, the systems that rely on them have to provide a corresponding level of assurance.”
– Henry Patishman, Executive Vice President of Identity Verification Solutions
Identity assurance has to continue beyond login
The solution is not to make every interaction repeat the entire identity verification process. Instead, organizations need to treat trust as something that can change throughout an identity lifecycle.
A routine action performed from a familiar environment may require no additional friction. But a sudden device change, account recovery, credential reset, privilege escalation, access to particularly sensitive data, or an unusual transaction can justify fresh evidence that the actor is still the legitimate identity holder.
This is particularly important as organizations begin interacting not only with humans and conventional bots, but also with legitimate AI agents acting on users’ behalf. In this new reality, the security challenge goes beyond simply distinguishing a human from a machine. It is establishing who or what is acting, whose authority it is using, and whether the action remains consistent with that authority.
For a deeper look at why a trusted credential does not automatically make the entire transaction trustworthy, read Henry Patishman’s article Why a Strong Credential Is Only the Start of the Trust Chain published in Biometric Update.
To explore how organizations can extend identity assurance beyond onboarding through ongoing monitoring, risk-based checks, and reverification, see our comprehensive guide on Identity lifecycle management.
.webp)
