VICTORIA, Seychelles, March 20, 2026 : Bitget, the world’s largest Universal Exchange (UEX), has collaborated with blockchain security firm SlowMist, to release a joint research report examining the risks emerging as AI systems begin executing trades autonomously. As trading enters this “agentic” phase, where systems move beyond analysis into action, a new category of risk is taking shape, one that traditional security models were not designed to address.
The report highlights a fundamental shift that once AI moves from advisory roles into execution, errors and exploits are no longer isolated events. They can trigger immediate, irreversible financial outcomes. In crypto markets, where transactions settle instantly, a compromised or misdirected agent can act faster than human intervention allows.
“AI is no longer just interpreting markets, it’s participating,” said Gracy Chen, CEO of Bitget. “That changes the nature of risk entirely. The question is no longer how intelligent these systems are, but how safely they are allowed to operate.”
According to the research, agent-based systems introduce new attack surfaces across multiple layers, from model inputs to execution pathways. Prompt injection can influence decision-making, malicious plugins can alter behavior and over-permissioned APIs can expose capital to unintended actions. These risks are compounded by the always-on nature of autonomous agents, which operate continuously without direct user oversight.
Rather than treating these as isolated vulnerabilities, the report frames them as systemic. Security in the agentic era must extend beyond application-level safeguards into the architecture of how AI systems interact with capital.
Bitget’s approach reflects this shift. The platform separates intelligence, execution, and asset authorization into distinct layers, reducing the likelihood that any single point of failure can trigger unintended trades. Permissions are structured around least-privilege access, with transaction simulation and verification processes introduced before execution is finalized. These controls are designed to ensure that even as AI agents operate autonomously, their scope remains defined and constrained.
SlowMist’s analysis reinforces the need for a closed-loop security model, where risks are addressed before, during, and after execution. Continuous monitoring, bounded permissions, and verifiable transaction flows form the foundation of this framework, moving security from a reactive process to an embedded system design.
The findings point to a broader reality where AI agents become more integrated into trading, asset management, and on-chain activity; the boundary between user intent and system execution becomes increasingly abstract. In this environment, reliability is no longer determined solely by performance, but by how well systems can operate within controlled limits.
Within Bitget’s UEX model, where crypto assets, derivatives, and tokenized traditional instruments coexist, this shift carries wider implications. As financial activity becomes more automated and interconnected, infrastructure must be designed not only for speed and access, but for containment and resilience. The joint report serves as a reference point for platforms, developers, and users navigating this transition, highlighting that the next phase of financial innovation will depend as much on secure execution as it does on intelligent systems.
To read the full report, visit here.
