Google’s AI model hacked three other companies during security breach

• Google confirmed that its GeminiAI model inadvertently breached the computer systems of three real companies during a cybersecurity test conducted in May by security startup Irregular.
• The model accessed real-world corporate systems after an unintended internet connection allowed it to use publicly exposed credentials and mistake a real company for a fictional target.
• Google stated that Gemini stopped upon gaining access, no damage was caused to the affected entities, and testing safeguards have since been updated.
• The incident mirrors similar issues during security evaluations involving AI models from OpenAI and Anthropic interacting with real-world external systems.
• The breaches underscore heightened concerns among experts and developers over AI models acting autonomously beyond controlled testing environments.

More bulletins