In September 2026, Spain's data protection agency (AEPD) reported the first recorded cyberattack carried out autonomously by an AI agent. The agent got into the organization's system, looked for weaknesses, changed users' personal data and went through invoices and financial reports. Neither the organization nor the software has been named. According to AEPD spokesperson Francisco Pérez Bes, a "well-known large language model" was used, and its developer was not compromised.
Not an isolated case
In a Digitalk article, Alexander Glavchev pulls together several signals:
- 2024, University of Illinois: GPT-4-based agents independently exploited 87% of 15 zero-day vulnerabilities in a test environment.
- Google Threat Intelligence Group: a financially motivated group used a multi-agent system combining models such as Gemini, Claude and Codex to compromise cloud infrastructure. According to the report, the agents harvested thousands of passwords in under 6 hours and rotated their IP addresses automatically.
- Speed of response: Hristo Laskov of Centio notes that his company saw a failed attack roughly six hours after a vulnerability was publicly disclosed.
Why speed changes everything
When an attack begins within hours of a vulnerability being published, an annual penetration test and manual updates "when there is time" are no longer enough. Laskov also warns about something else: AI removes the typical mistakes from phishing emails and makes it possible to send personalized messages built from leaked data.
What the experts recommend
- Strong baseline controls.
- Faster vulnerability management.
- Identity protection.
- Oversight of suppliers.
- Strict control over the organization's own AI agents.
- Realistic security testing instead of relying on annual tests alone.
What this means for WordPress websites
For most websites, "faster vulnerability management" means something concrete: automatic or at least weekly plugin updates, removing unused plugins, protecting the login and monitoring. Maintenance is not a luxury; it is part of your defense. See what WordPress maintenance and security includes.
Related: the AISI tests with AI agents and five lessons from the Revolut breach.
Topic source: Digitalk: Autonomous AI agents are the new reality in cybersecurity (in Bulgarian). This article is an independent analysis by Penkov Studio.
