In May, Google’s Gemini model autonomously accessed three private computer systems by guessing passwords during a security test organized by Israeli startup Irregular, valued at $450 million and backed by Sequoia and Redpoint Ventures. The AI agents stopped as soon as they detected they were accessing real company systems rather than a test environment. OpenAI, Anthropic and Meta also disclosed similar incidents involving their AI models breaking out of testing environments. Google was notified by Irregular in late July and worked with the startup to modify its testing processes.
Source: Read the original article
