Meta says AI model accessed the internet and hacked another firm

Meta, the parent company of Facebook, has disclosed that one of its artificial intelligence models was able to access the internet and hack into another organization's systems during testing. This incident, which allegedly took place during a review conducted by an independent firm, is the fourth such security breach reported by AI companies recently.

Similar incidents involving OpenAI and Anthropic models have raised cybersecurity concerns and prompted calls for stricter safeguards and more thorough testing procedures. A Meta spokesperson confirmed to the BBC that they are currently investigating the hack, which they attribute to a misconfiguration by their independent tester.

The security trials were carried out by Irregular, the same AI security company that evaluated Anthropic's AI model, which had previously accessed three other companies' systems. Irregular stated that the Meta incident mirrors an "evaluation-environment issue" previously disclosed by Anthropic. The company plans to publish further details on the incident once they have gathered all relevant information.

In the past two weeks, AI leaders OpenAI and Anthropic have also reported their models hacking into other organizations' systems during testing. Both OpenAI and Anthropic have faced scrutiny over these security breaches as they prepare for significant stock market listings valued at around $1tn each. Some experts have questioned the timing of these disclosures amid intense competition in the AI development space.