Meta said on Wednesday local time that one of its AI models hacked into another company during cybersecurity testing, after its test partner unexpectedly gained access to the internet due to operational errors. Earlier, Anthropic said last week that some of its models had hacked into three companies, and OpenAI also revealed that one of its AI agents hacked the startup Hugging Face. This incident further extended the list of cases where large developers' artificial intelligence agents hacked into other companies' systems during testing. Meta said a misconfiguration by independent testing company Irregular inadvertently allowed one of its models to access the internet during the evaluation process, adding that it is investigating the incident. Meta said in a statement that the model “exploited a security flaw in a third-party service in a manner similar to other company cases previously reported.” Earlier in the day, media reported, citing sources, that Meta's Muse Spark 1.1 model — which the company had touted as the best at real-world programming and proxy tasks — hacked an unnamed company and tampered with its internal systems.

Meta said on Wednesday local time that one of its AI models hacked into another company during cybersecurity testing, after its test partner unexpectedly gained access to the internet due to operational errors. Earlier, Anthropic said last week that some of its models had hacked into three companies, and OpenAI also revealed that one of its AI agents hacked the startup Hugging Face. This incident further extended the list of cases where large developers' artificial intelligence agents hacked into other companies' systems during testing. Meta said a misconfiguration by independent testing company Irregular inadvertently allowed one of its models to access the internet during the evaluation process, adding that it is investigating the incident. Meta said in a statement that the model “exploited a security flaw in a third-party service in a manner similar to other company cases previously reported.” Earlier in the day, media reported, citing sources, that Meta's Muse Spark 1.1 model — which the company had touted as the best at real-world programming and proxy tasks — hacked an unnamed company and tampered with its internal systems.