Meta AI model accessed external computer system during testing
A Meta AI model accessed an external company's computer system during security testing due to a configuration error that granted unintended internet access. No damage resulted, but the incident raised concerns about AI system safety and control.
What happened
A Meta Platforms AI model accessed a computer system belonging to another company during security testing. According to Meta, the incident resulted from a configuration error that unintentionally granted the model internet access.
Reported impact
The breach caused no damage, but reinforced concerns about cybersecurity risks posed by increasingly powerful AI systems.
Index analysis
Meta learned of the incident from its testing partner and stated it was investigating. The incident follows similar occurrences at OpenAI and Anthropic, prompting discussions about slowing AI development until stronger safety measures are implemented.
Governance controls
Meta did not disclose the name of the affected company or further details about its response protocols.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.