KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
Regulation
Neutral

AI Agents Expand Role in Web3 Security Monitoring and Response

Published October 7, 2026 2:26 PM · 6 views
AI Agents Expand Role in Web3 Security Monitoring and Response

AI agents are expanding beyond acting as auxiliary tools for classifying alerts in Web3 security work to becoming active entities that monitor transactions and halt smart contract functionality. A proposal has emerged suggesting that the scope and authority of agents should be defined, with critical decisions and actions remaining under human supervision.

CertiK's October 5 report 'The Rise of the AI Security Workforce' proposes managing AI agents as part of an organization's operational structure, outlining their roles, permissions, reporting lines, and responsible personnel. The report explains that agents can investigate alerts through multiple stages, gather evidence from external tools, and take actions within assigned permissions.

Web3 security examples include smart contract vulnerability checks and on-chain transaction monitoring. If suspicious transactions or exploit indicators are detected, supported contract functions may be temporarily suspended or circuit breakers activated. However, the report emphasizes that actual risk assessment of vulnerabilities and scope of audits must be confirmed by humans.

In anti-money laundering (AML) work, agents can track fund flows across multiple addresses and bridges, analyze address correlations, and draft regulatory reports. The responsibility for verifying evidence and approving report content remains with people. If agents hold or trade digital assets, they should document the information used for decisions and actions taken, creating audit trails to verify post-hoc.

CertiK notes that automation carries risks of misjudgment and manipulation. Models may incorrectly classify threats or produce inaccurate reports, and manipulated inputs or prompt injections could distort the behavior of agents with significant authority. Attackers can also leverage AI for vulnerability scanning and social engineering attacks. As agent execution authority increases, so does both the speed of defense work and potential impact of incorrect judgments.

CertiK recommends maintaining audit logs for all key inputs, decisions, and actions; pre-defining human approval requirements for certain activities; conducting regular checks against adversarial manipulation; and clearly assigning responsibility. The report adds that how to allocate accountability in cases of frozen transactions, inaccurate reports, or incorrect contract halts remains unresolved across jurisdictions.

The trend of granting execution authority to AI agents is also visible in trading operations, as previously reported by this publication regarding Binance opening order permissions to AI agents. CertiK's report suggests that when applying autonomous execution to security, AML, and compliance tasks, permission scope, approval procedures, and post-hoc documentation should be designed together.

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: AI 에이전트, 웹3 보안 감시·대응까지 맡는다