Blockchain security firm CertiK says agentic AI is pushing cybersecurity, anti-money laundering (AML) and compliance toward increasingly autonomous operations, with human professionals shifting from routine investigative work to supervising AI-driven systems. The company laid out the trend in a new report, "The Rise of the AI Security Workforce: How Agentic AI Is Redefining Cybersecurity, AML & Compliance," which examines how autonomous AI systems are transforming security and compliance functions across both Web2 and Web3.

Unlike conventional AI tools, agentic AI can handle multi-step processes, retrieve and synthesize information, interact with external systems, and take actions with little or no human intervention, according to CertiK. In a cybersecurity scenario, an agent could investigate an anomalous login, correlate device and threat-intelligence data, conclude that an account poses a risk and suspend it, while maintaining a record of every action it takes.

The report identifies three forces driving adoption: the accelerating speed of cyberattacks and money laundering, persistent shortages of cybersecurity and financial-crime professionals, and expanding regulatory requirements. CertiK notes that AML penalties exceeded $900 million in the first half of 2025, adding to pressure on organizations to scale compliance operations without relying solely on additional personnel. Taken together, the pressures describe a capacity gap — accelerating threats and obligations on one side, a persistently short supply of specialists on the other — that helps explain why organizations are looking to automate oversight rather than simply expand headcount.

In traditional Web2 environments, agentic systems can autonomously triage alerts, assess vulnerabilities, monitor identity and access behavior, and continuously check systems against cybersecurity controls. In AML workflows, they can re-evaluate transaction alerts, conduct customer due diligence, build relationship graphs and prepare suspicious activity reports for human review. Compliance teams can also deploy AI agents to monitor regulatory changes, identify affected internal policies and perform continuous audits rather than relying on periodic reviews. The throughline across these examples is a move from periodic review toward continuous operation, with humans reviewing agent output rather than executing each step themselves.

Web3 Applications Raise Both Capabilities and Risks

CertiK identifies Web3 as an area where agentic AI can have a particularly significant impact, because blockchain transactions are irreversible and security responses often need to occur within minutes or even seconds. AI systems can assist with smart-contract auditing, detect exploit patterns in real time and, in advanced deployments, trigger protective measures such as contract pauses or circuit breakers. They can also reconstruct complex attack paths after an exploit and generate technical post-mortems.

The report additionally highlights autonomous on-chain AML capabilities, including real-time fund tracing, dynamic address clustering and cross-chain analysis. Compliance systems are increasingly moving toward pre-settlement Know-Your-Address and Know-Your-Transaction risk scoring, allowing potential exposure to illicit activity to be identified before transactions are completed. The practical effect is that risk surfaces before settlement rather than after funds have already moved.

At the same time, CertiK warns that greater autonomy introduces new risks. AI agents can produce incorrect assessments, attackers can exploit or manipulate agentic systems, and questions around liability remain unresolved when autonomous actions cause harm. The report recommends defined authority limits, comprehensive audit trails, regular adversarial testing and a named human owner responsible for each AI agent.

CertiK ultimately argues that organizations should treat agentic AI as a workforce role rather than simply another software tool. As AI systems take on more operational responsibilities, overseeing their behavior and preserving evidence of their decisions is itself emerging as a distinct compliance function. For readers tracking where this goes, the report's own unresolved points — how liability is assigned when autonomous actions cause harm, and whether the recommended guardrails become standard practice — are the concrete developments to watch.

This article was produced from reporting originally published by Metaverse Post.