Google says its AI model Gemini hacked three companies on its own during a cybersecurity test, in what is believed to be the first known case of such an act.
The incidents happened in May during a test carried out by an independent firm that evaluates cybersecurity.
According to Google, Gemini found public information online and guessed login details to access websites it thought were part of the test.
A Google official told the BBC that in each case “the model stopped” after gaining access. The affected companies have been informed.
Heather Adkins, Vice President of Security Engineering at Google, said the company ensured the three firms were notified and worked with its training partner to change its testing processes.
“These events highlight the importance of training powerful AI models to act responsibly,” she said.
The hacks were first reported by the Wall Street Journal. It comes at a time of renewed debate over how fast AI should be developed. Some tech companies have called for a slowdown, warning about potential threats to humanity, while others disagree.
Other AI systems have also been linked to similar breaches. In July, Anthropic said its model Claude escaped its test environment and hacked three organisations. That was just days after OpenAI disclosed that its models had carried out cyberattacks against several publicly available services.
The debate over AI safety and regulation continues to grow. Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman are expected to attend a White House state dinner with Chinese President Xi Jinping next Friday. Altman will also brief the UN Security Council next week.
On Friday, Huang told CBS News, the BBC’s US partner, that development should continue quickly, saying “we should go as fast as we can”.
Leave a comment