Security Researcher Confirms 'Poison Nest' Backdoor Issues in Trae AI Plugin Market

Security researcher Cos (@evilcos) has confirmed a serious 'Poison Nest' backdoor problem in Trae AI's plugin marketplace. The term refers to a cluster of malicious plugins containing backdoors that demonstrate remarkable resilience, continuously receiving updates and persisting on the marketplace despite detection efforts.

Trae AI has been widely used by developers as an AI agent tool, with its plugin marketplace hosting numerous third-party extensions. According to Cos's findings, these backdoor plugins expose developers and regular users to severe security risks including data theft and remote system control.

This incident represents yet another major supply chain security breach in the AI industry. Browser extensions and open-source tools have historically been targeted for backdoor implants, but AI agent plugins amplify the risk significantly — they can access user context, read file systems, and even execute code, making the potential damage far greater than traditional software plugins.

The Trae platform now faces a crisis of trust. Key questions remain: whether the marketplace's review mechanism is robust enough, whether identified backdoor plugins have been removed, and how the platform will address its security vulnerabilities.

Industry-wide, this event could push AI plugin markets toward either centralized auditing or decentralized verification models. Security audit tools and trusted plugin platforms are likely to attract increased capital and user demand, while current plugin supply chain standards face pressure to accelerate formalization.

The AI agent ecosystem remains in its early stages of security governance. Similar backdoor incidents may emerge on other AI platforms, making effective plugin security review mechanisms and the balance between openness and safety a long-term challenge for the entire industry.