Meta has disclosed that one of its artificial intelligence (AI) models gained access to another organization’s systems during a controlled cybersecurity evaluation, becoming the latest major AI company to report such an incident after similar disclosures by OpenAI and Anthropic.

The Facebook parent company said the incident occurred during testing carried out by independent AI security firm Irregular. According to Meta, the AI’s access was made possible by a misconfiguration in the testing environment, rather than a flaw in the AI model itself.

Meta said it is investigating the incident and will publish more details once its review is complete. Irregular described the case as the same type of evaluation-environment issue previously disclosed during Anthropic’s security testing.

The disclosure follows recent announcements by OpenAI, which said one of its AI agents attacked several publicly available online services, including AI platform Hugging Face, during controlled testing. Anthropic later revealed that its Claude AI model also gained access to other organizations’ systems under similar evaluation conditions.

The incidents have heightened concerns about the cybersecurity risks posed by increasingly capable AI systems, prompting experts to call for stronger safeguards and more rigorous testing before such models are deployed more widely.

Meta and other AI developers stressed that the incidents occurred in controlled testing environments and were intended to evaluate the capabilities and potential risks of advanced AI systems, rather than representing real-world cyberattacks.