TeamPCP Demands $25,000 for Nearly 5 GB of Stolen Mistral AI Code, Threatens Free Leak

Hackers operating under the name TeamPCP are attempting to sell a substantial cache of proprietary Mistral AI source code, demanding $25,000 in exchange for roughly 5 gigabytes of stolen material. The group has set a one-week deadline, warning that if no buyer is found, it will release the data for free. The breach, if confirmed as described, ranks among the more significant intellectual property thefts targeting a frontier AI laboratory in recent memory.

The attackers claim to have exfiltrated approximately 450 internal repositories belonging to Mistral AI, one of Europe's most prominent AI developers. Among the stolen assets, the group says it obtained sensitive source code used for model training and model delivery pipelines. That specific category of code gives the stolen material particular sensitivity: training infrastructure code can reveal proprietary methodologies, while delivery systems code may expose deployment architectures and security design. Mistral AI has not issued a public statement confirming or denying the breach, and the authenticity of the stolen data has not been independently verified as of this writing.

The incident places immediate pressure on Mistral AI to assess the scope of a potential compromise and determine whether the stolen code represents a genuine threat to its competitive position or to the security of systems built on its models. For the broader AI sector, the episode underscores the escalating interest of threat actors in stealing proprietary AI assets, where even partial code leaks can provide competitors or malicious actors with insight into training techniques or expose unpatched vulnerabilities. The one-week deadline introduces a narrow window: absent a buyer or an official response that neutralizes the threat, the attackers appear prepared to maximize exposure of the stolen material.