Artificial Intelligence (AI) goes beyond recommendations to execute transactions, and errors or security breaches can have immediate financial effects, according to a report by Bitget and SlowMist.
With AI's migration from recommendation to action, a new category of risk emerges for which traditional security models have not yet been designed.
„When Artificial Intelligence goes beyond recommendations to execute trades, errors or security breaches can have immediate financial effects. In the crypto market, where transactions are processed almost instantaneously, a compromised or misdirected agent can make decisions before humans can intervene. Autonomous AI systems open new security vulnerabilities, from the data they receive to the way they execute transactions. Specifically, these systems can be influenced by misleading commands, malicious apps or overly permissive APIs, which can lead to unwanted actions. Thus, because these types of agents run continuously, without user oversight, the risk of an error or attack causing negative effects is even higher”, says the report, according to Agerpres.
Experts say these vulnerabilities are not isolated cases, but part of a systemic risk.
„So, in this new phase, security can no longer be thought of only as safeguards at the application level, but must be integrated into the entire architecture through which AI systems interact with the capital. Against the backdrop of these risks, platforms that use AI agents are having to rethink the security measures adopted, the report highlights. One of the solutions increasingly being advocated is a clear separation between analysis, execution and access to funds, to avoid situations where a single vulnerability can trigger unintended transactions. Permissions are structured on the principle of minimum necessary access, and transaction simulation and verification processes are introduced before execution is finalized. These controls are designed to ensure that although AI agents operate autonomously, their scope of action remains clearly defined and limited”, notes the source.
According to the report, there is a need for a security model that works at all stages of a transaction, i.e. before, during and after execution.
At the same time, continuous monitoring, limited permissions and verification of each transaction are basic elements in such a framework, and security is no longer just a reaction, but an integrated system.
„The data points to a broader reality where AI agents are becoming increasingly embedded in trading, asset management and on-chain operations, and the boundary between user intent and system execution is becoming more abstract. In this context, trust is no longer determined solely by performance, but also by the ability of systems to operate within controlled boundaries. Financial activity is becoming increasingly automated, but infrastructure must be designed not just for speed and access, but also for control and resilience”say experts.
The report by Bitget and SlowMist is a benchmark for digital asset trading platforms, developers and users navigating this transition.