Google's Gemini AI successfully breaches corporate websites in security trial

Google has revealed that its Gemini artificial intelligence model successfully penetrated the defenses of three separate companies during an authorized security evaluation, according to statements made to the BBC by a company official.

The AI system operated autonomously to access the internet and systematically guessed login credentials to gain entry into the targeted websites. The breach attempts were conducted with the knowledge and permission of the companies involved, forming part of a deliberate assessment of security vulnerabilities and AI system capabilities.

The exercise underscores the growing sophistication of large language models and their ability to perform multi-step digital tasks independently. Researchers and security experts have increasingly focused on understanding how advanced AI systems might be weaponized or exploited, particularly regarding their capacity to operate without direct human intervention.

Google's demonstration comes amid broader industry discussions about responsible AI development and the importance of rigorous security testing before deployment of powerful new technologies. The trial results are expected to inform ongoing conversations about safeguards and ethical frameworks for artificial intelligence systems as they become more capable and widespread.