Meta, Facebook’s parent company, is the latest tech business to admit one of its AI models managed to connect to the internet and hack into another organisation’s systems during testing.
Meta said the incident happened during a review by an independent organization. It is the fourth recent instance of its kind acknowledged by AI companies. Cybersecurity worries and calls for stronger controls and more thorough testing have been fueled by similar breaches from OpenAI and Anthropic models.
A Meta representative told the BBC it was examining the attack, which it said was triggered by a “misconfiguration” by its independent tester. It also said the incidence was comparable to what had been reported at other organizations in the past.
Meta stated the security testing were undertaken by Irregular, the same AI security provider that tested Anthropic’s AI model that had breached three other companies’ systems.
The Meta incident is the very same evaluation-environment issue already disclosed by Anthropic last week,” an Irregular spokeswoman said. The firm’s representative told the BBC Irregular is working on a paper on how to execute cyber-security testing involving AI agents securely.
Also Read:
China’s Economic Growth Drastically Declines, Falling Short of Expectations
Public Ownership of British Steel to Safeguard Vital” UK Supplies
