The AI aggregator OpenRouter has revoked the Zero Data Retention (ZDR) designation, which guarantees that no requests or responses are stored, for the affected MiniMax models. The formal trigger was a study by cybersecurity specialists at Dragos, during which alarming signs of a leak were identified: unique links embedded in test requests remained accessible 40–130 hours after being sent.
Such infrastructure behavior directly contradicts the principles of ZDR and indicates possible third-party access to the contents of the correspondence. For users who trusted the privacy status when working with sensitive data, this is a signal of the need to reconsider their policy of interaction with a number of providers.
The scope of the audit went beyond a single model
As part of a broader audit, OpenRouter removed 125 endpoints from the ZDR list and completely disabled three providers that failed to ensure the proper level of data protection. The company also notified potentially affected clients — a step that confirms the seriousness of the identified violations and points to the systemic rather than isolated nature of the problem.
The situation exposes a fundamental vulnerability of aggregator models of access to AI: the user often does not control which infrastructure their request actually passes through. Even the presence of a formal ZDR designation does not guarantee compliance with the stated obligations at the level of individual providers within the processing chain.
For the market, this is another reminder that privacy in the generative AI segment remains a zone of heightened risk. Corporate clients working with confidential data should require independent verification of retention policies rather than relying solely on platform declarations.
Cryptalist analysis: The incident increases pressure on AI model aggregators regarding the transparency of data processing chains and may accelerate the corporate segment's shift toward providers with verifiable ZDR guarantees. For the MiniMax ecosystem, the reputational costs will prove more sensitive than the technical ones, while the market itself will receive an incentive to standardize privacy auditing as a competitive advantage.