Certik's 'Agentic AI' Redefines Security and Compliance Workflows

Certik, the global leader in Web3 security, released its October 5 Intel3D report titled "The Rise of the AI Security Workforce: How Agentic AI Is Redefining Cybersecurity, AML, and Compliance." The report analyzes how AI is evolving from an analytical assistant to a team member capable of conducting investigations, making judgments, and taking actions within limited permissions.
This shift is reshaping security and compliance workflows in both traditional finance and Web3 sectors while presenting new challenges for enterprises regarding how to supervise AI, define permission boundaries, and assign accountability for outcomes. The report cites accelerating attack speeds, talent shortages, and increasingly stringent regulatory demands as key drivers of this change.
In the digital asset space, attacks can occur through a single transaction with stolen funds rapidly moving across multiple addresses and cross-chain bridges within short timeframes. Meanwhile, security and compliance teams face growing volumes of alerts and transactions that simple headcount increases cannot address in real-time monitoring and investigation needs.
Differing from conventional AI tools that merely flag anomalies or generate text, Agentic AI performs multi-step reasoning, calls external tools, synthesizes evidence from different sources, and determines next actions based on outcomes. For example, upon detecting an unusual login, it can synthesize device information, historical login locations, and threat intelligence to conduct investigations, take predefined actions within set permissions, and leave records for human review.
This transformation is evident in existing security audits and AML investigations. Certik states Agentic systems can correlate logs across different systems to initially categorize alerts and forward cases requiring additional verification to responsible parties. In AML investigations, it synthesizes transaction histories, customer information, and relationships between entities to draft risk assessments and suspicious activity reports.
Compliance roles are increasingly focusing on reviewing evidence, handling complex cases, and making final decisions on reporting. In the Web3 sector, smart contract security audits and on-chain fund tracking are highlighted as key use cases. AI-based audit agents can analyze call relationships and state changes across multiple smart contracts to identify issues like access control misconfigurations or unsafe upgrade designs.
According to Certik's real-world security audit experience, the role of skilled auditors is also evolving. There is growing importance in verifying AI-detected issues, analyzing complex attack paths, and assessing the detection capabilities and blind spots of AI tools. However, final security judgments remain with engineers.
The expansion of cross-chain fund movement has also changed investigation approaches. The report cites Certik's prior research showing that 86% of stolen ETH after the Bybit hack was converted to Bitcoin within a month, utilizing mixing services, cross-chain bridges, and over-the-counter (OTC) transactions. To track such complex fund flows, continuous connection of clues across multiple blockchains is required. Here, Agentic AI goes beyond reconstructing transaction paths post-incident to continuously advance investigations based on fund movements.
As AI enters compliance workflows, the scope has expanded from post-transaction investigation to pre-transaction and routine operations. AI agents can evaluate counterpart addresses and transaction risks before payments or transactions occur, continuously check whether system settings meet internal controls and regulatory requirements, and assist in drafting regulatory reports.
This allows compliance teams to detect risks earlier and reduce the time between anomaly detection and human intervention. However, as AI agents begin to hold, trade digital assets, or execute trading strategies, monitoring AI behavior itself has become a new compliance challenge. Enterprises must audit AI's on-chain activities, maintain records of data used by AI, its reasoning basis, and actual tasks performed for future review and forensic analysis.
The report emphasizes that while AI performs compliance work, determining whether AI itself complies with regulations is a separate issue requiring organizations to still bear responsibility for AI agent behavior. Autonomous execution capabilities increase efficiency but also amplify error impacts. Certik's report warns of potential AI misclassifying actual breaches as normal activity, generating incorrect fund movement paths, or wrongly assessing smart contract security.
Attackers can also leverage AI to more rapidly conduct vulnerability scanning, social engineering attacks, and reconnaissance activities. Security and compliance AI agents with access to sensitive data and the ability to call external tools are themselves potential targets. Threats include prompt injection or input manipulation leading to AI approving fraudulent transactions or disabling normal controls.
To address these risks, Certik recommends enterprises clearly define each agent's role, permissions, and escalation procedures from the outset of deployment. Distinguish between tasks AI can autonomously perform versus those requiring human approval, maintain complete audit trails for key decisions, and conduct regular red team testing. Assign clear human accountability to each AI agent for performance and errors.
While AI's scope of responsibility will expand, organizations remain accountable for outcomes. The widespread adoption of Agentic AI presents enterprises with the dual challenge of technology implementation and governance framework development. Organizations lacking sufficient capability may fall behind in processing speed, cost efficiency, and response scope, while those deploying autonomous systems without proper oversight face new operational and compliance risks.
The rise of "AI employees" is transforming security and compliance workflows, and whether enterprises can sustainably leverage these benefits depends on how effectively they develop governance capabilities to match AI's execution capabilities.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 서틱 "에이전틱 AI가 보안·컴플라이언스 업무 재정의"