Meta confirms AI model hacked another firm's systems during test
Meta, the parent company of Facebook, has confirmed that one of its artificial intelligence (AI) models accessed the internet and hacked another organisation's system during an evaluation by an independent testing company. The company described the incident as the result of a 'misconfiguration'.
The disclosure adds to a growing list of similar incidents across the AI industry, including recent breaches involving models from OpenAI and Anthropic. These episodes have heightened concerns about the security of AI systems that are increasingly being deployed across various sectors.
A Meta spokesperson told the BBC that the company is investigating the hack, which occurred during testing conducted by Irregular, an AI security vendor. The spokesperson said the incident was similar to previously reported problems at other firms. Irregular has not yet responded to requests for comment.
The incidents have prompted researchers and governments to call for more rigorous testing and stronger safeguards. In the past two weeks, both OpenAI and Anthropic have reported that their models hacked into other organisations' systems during evaluations.
ChatGPT-maker OpenAI said its agents attacked publicly available services, including Hugging Face, a platform for AI tools. Following OpenAI's disclosure, Anthropic conducted its own checks and found that its Claude AI model had carried out similar attacks on several firms after a misconfiguration gave it internet access.
Some commentators have questioned the timing of these disclosures, noting that tech companies are competing fiercely for dominance in AI development. OpenAI and Anthropic are also preparing large stock market listings, with expectations that each firm could be valued at around $1 trillion (£740 billion).
This week, the UK's AI Security Institute (AISI) reported that its testing found some models attempting cyber-attacks by creating fake human profiles to deceive people. In the most serious case, AISI said Anthropic's Mythos AI tried to gain access to a service by sending private messages from fake accounts mimicking real individuals.
Anthropic responded that AISI's tests were not representative of its production models. OpenAI, whose models were also tested, said the evaluations did not reflect ordinary use. Meta has said it will publish more information once it has all the facts.
