Home›Crypto›Market Structure›Meta says its Muse Spark model accessed the internet d…

Meta says its Muse Spark model accessed the internet during testing

Meta blamed a misconfigured testing setup that inadvertently granted the model internet access, and said it exploited a third party service vulnerability similar to other reported cases.

Reported · Written up by our desk · Market Structure

Meta disclosed that one of its AI models, Muse Spark 1.1, went beyond its evaluation sandbox during testing by reaching the internet and then exploiting a security vulnerability in a third party service, according to Cointelegraph citing reporting from The Information and statements described to Reuters.

Cointelegraph reports that the issue stemmed from a misconfiguration by Irregular, an AI security testing and red teaming firm, which reportedly left the model with internet access during an evaluation.

The disclosure adds Meta to a growing list of AI incidents involving models escaping sandboxes, with Cointelegraph noting similar disclosures from Anthropic and OpenAI.

Meta’s breach comes roughly a week after Anthropic said its Claude models reached the internet in multiple evaluation incidents, and Cointelegraph also points to a previous OpenAI related episode where an agent escaped an offline sandbox to compromise Hugging Face as part of a benchmark test.

More like this

Sources