KI-Angriffe: BSI-Präsidentin Plattner: „Wir reden von durchaus dramatischen Szenarien“

Frankfurt and Düsseldorf – A security breach did not occur, but a warning was issued. During the winter, an anonymous hacker group gained access to the IT systems of a water utility in Monterrey, Mexico. The intruders used artificial intelligence (AI) to automate many work processes, including writing 17,000 lines of program code using a model from Anthropic.

The US cybersecurity firm Dragos documented this, noting that the AI system identified an interface in the network as a "desirable target" and attempted to gain access. However, this attempt was unsuccessful as the control systems governing the water infrastructure were not compromised. Despite the favorable outcome, the incident demonstrates how AI is changing and exacerbating the cybersecurity landscape for businesses, particularly those managing critical infrastructure.

The pressure to act is clearly higher following this event in Mexico. Models like Claude Mythos have proven effective in finding security vulnerabilities. Additionally, a swarm of AI agents from OpenAI recently autonomously hacked the developer platform Hugging Face without a specific directive. The business community is now alarmed, with the threat from autonomous attacks being a top concern in executive suites and crisis rooms of banks and other critical infrastructure operators.

While most want to remain silent on this sensitive topic, even leaders of major AI companies are warning that their systems may be out of control. Claudia Plattner, President of the Federal Office for Information Security (BSI), emphasized in a video message that "we are talking about very dramatic scenarios that we need to be prepared for," noting that even if the AI had no malicious intent, it was still fulfilling its programmed task.