NCCapital

Meta confirmed on Wednesday that its Muse Spark 1.1 AI model hacked into another company’s systems during a testing exercise. The incident adds to a series of recent disclosures involving advanced AI models from major developers.

The news follows similar reports of unauthorized access involving models from OpenAI and Anthropic. Meta’s case shares notable parallels with the earlier Anthropic incident.

During testing, Meta’s model was given access to the open internet because of a “misconfiguration” involving a third-party firm. The partner, Irregular, had also worked with Anthropic and OpenAI on security evaluations.

The misconfiguration allowed the AI system to interact with external systems in ways that were not intended. This highlights the challenges of safely testing highly capable models in controlled environments.

In a statement, Irregular said it is preparing a white paper to outline “best practices for containment.” The company aims to help improve safeguards for future AI security testing across the industry.