Bitget, the world’s largest Universal Exchange (UEX), has teamed up with blockchain security firm SlowMist to release a comprehensive research report that explores the security risks linked to the increasing use of artificial intelligence (AI) in trading activities. As trading systems begin to operate in what the report describes as an “agentic” phase—where AI goes from merely analyzing data to executing trades autonomously—traditional security frameworks are proving inadequate for addressing the newly emerging risks.
The report emphasizes a crucial transition: once AI evolves beyond just providing advice, the consequences of errors or exploits can lead to immediate and irreversible financial disasters. This rapid action capability is particularly concerning in cryptocurrency markets, where transactions finalize almost instantaneously, making it difficult for human oversight to intervene effectively.
Gracy Chen, CEO of Bitget, remarked on the implications of this evolution, stating, “AI is no longer just interpreting markets; it’s participating. That changes the nature of risk entirely.” This new environment raises concerns not just about the intelligence of AI systems, but about ensuring their operations remain safe and controlled.
Key findings from the research indicate that these agent-based systems open up new attack surfaces at various levels, including model inputs and execution pathways. Issues such as prompt injection—which can skew decision-making—malicious plugins that alter system behavior, and over-permissioned APIs that expose assets to potential misuse contribute to this risk landscape. Furthermore, the continuous operational nature of autonomous agents exacerbates the vulnerability, as they function without direct user oversight.
The report advocates recognizing these vulnerabilities as systemic rather than isolated, suggesting that security measures must encompass a broader architectural approach to how AI systems interact with capital. In response to these findings, Bitget has adapted its security framework to delineate clearly between intelligence, execution, and asset authorization into separate layers. This structure minimizes the risk that a single point of failure can lead to unintended transactions. By employing principles of least-privilege access and incorporating transaction simulation and verification processes, Bitget aims to ensure that AI agents can operate within confined boundaries, even when functioning autonomously.
SlowMist’s participation in this research underscores the need for a comprehensive security model that addresses risks at all stages—before, during, and after transaction execution. This proactive approach includes continuous monitoring, limited permissions, and verifiable transaction pathways, effectively shifting security from a reactive measure to an integrated aspect of system design.
The report also points to a future where AI agents become increasingly woven into the fabric of trading, asset management, and blockchain activities, leading to an increasingly blurred line between user intent and system execution. Consequently, the reliability of financial systems will hinge not just on performance metrics, but on the ability to operate within well-defined limits.
In Bitget’s UEX model, where a diverse range of assets—from cryptocurrencies to traditional financial instruments—intersect, the implications of this shift are profound. As financial activities become more automated and interconnected, there is a pressing need for infrastructure that is designed not only for speed and accessibility but also for containment and resilience. This collaborative report serves as a vital resource for platforms, developers, and users as they navigate this complex transition, reiterating that the next stage of financial innovation will rely equally on secure execution and advanced intelligent systems.
For those interested in exploring the complete findings, the full report is available on Bitget’s website.
