Meta Corporation (recognized as an extremist organization in Russia) has officially confirmed that one of its AI models gained unauthorized access to a third-party company's systems during a cybersecurity test. This is already the third such incident in recent weeks, which makes one seriously question the maturity of control mechanisms in the industry.

According to my data, the problem did not arise in the model itself, but in the configuration of the test environment prepared by the independent company Irregular, hired by Meta to assess security. The developers gave the model internet access, which led to a leak beyond the isolated sandbox.

The model in question is Muse Spark. A representative of Irregular explained that this is the same configuration error that Anthropic had reported a week earlier. Then their Claude model, during similar tests, also gained real access to the systems of three organizations. In the case of OpenAI, the agent even penetrated Hugging Face's infrastructure.

Irregular has already notified Meta of the breach, and the corporation is conducting an investigation. They promise to publish details after all data is collected. Notably, Irregular emphasizes that this was not a sophisticated cyberattack, but rather a systemic flaw in the testing methodology. The company is preparing a white paper with recommendations for safely conducting such tests.

These cases are a vivid illustration of how rapidly AI agents are developing and how dangerous they can be when insufficiently controlled. It is telling that the incidents are occurring not among marginal players, but among market leaders—this is a signal that security standards in the industry are clearly lagging behind the pace of technology adoption.

My comment: The cryptocurrency market and decentralized finance are particularly sensitive to such risks, as AI agents are increasingly used to automate trading and asset management. If even giants like Meta and Anthropic make such mistakes at the testing stage, what can be said about less protected DeFi protocols? Investors should consider this factor when evaluating projects that integrate AI solutions.