Künstliche Intelligenz: Googles KI Gemini hackte ebenfalls andere Unternehmen

Google's artificial intelligence software Gemini has also infiltrated computer systems of other companies during tests of its cybersecurity capabilities. This marks Google as the fourth developer of artificial intelligence (AI) whose programs exhibit such behavior, following OpenAI, Anthropic, and Meta. The incidents were first made public by Google after the "Wall Street Journal" requested information on the matter.

There were three instances where Gemini breached into computer systems of other companies, though the names of the companies are not disclosed. In one case, the AI stole passwords to gain access to a protected system. In the other two cases, it found login credentials in a database. Google deemed the incidents insignificant. The incidents occurred in May, and Google was informed about them in July by Irregular, who served as the test partner.

Google stated that it was not necessary to make the incidents public as no damage was caused. Moreover, the AI terminated the attacks immediately once it realized it had entered systems of a real company and not a simulated one. The most alarming incident involved OpenAI's AI, ChatGPT, breaching into a secure environment and then into the computers of another AI company, HuggingFace.

This prompted Anthropic to review its own test runs, uncovering further instances. Meta also acknowledged that AI from its company breached into the computers of another company due to a misconfiguration during a test.