AI-Powered Credential Theft Campaign Executed in Six Hours

As AI tools assist attackers in credential theft and vulnerability analysis, cryptocurrency companies must extend their security response beyond account protection to include transaction approval and fund movement controls. Google Threat Intelligence Group revealed in a September 9, 2026 report that attackers infiltrated an organization's cloud infrastructure and used AI coding tools and agent directives to plan, build, and execute a large-scale credential theft operation within six hours. The campaign targeted thousands of third-party credentials, though the time taken to gain cloud access was not included in the six-hour window. Google noted some attackers went beyond merely querying models by leveraging agent workflows and automation tools to minimize human judgment and processing delays.
The report also stated that attackers used AI models to convert exploited public vulnerabilities into code, repeatedly adjusting related elements. However, no fully autonomous attack flow targeting zero-day vulnerabilities against actual targets was observed. Google clarified that while AI has been used in attacks, this does not mean autonomous attacks on cryptocurrency firms are imminent. The confirmed change in the reported case is the use of AI tools for planning and executing credential theft after infiltration.
Crypto firms must be aware that employee accounts, developer accounts, and vendor systems can lead to fund permissions. If accounts are compromised or information displayed on transaction approval screens is manipulated, attackers may exploit approval processes without directly stealing wallet private keys. Bybit reported in February 2025 that credentials of Safe Wallet developers were stolen, with signers tricked into approving malicious transactions; the company stated forensic agencies found no intrusion traces within their own infrastructure.
Bybit's announcement did not cite AI use as evidence for this attack. If multiple signers see identical manipulated information, independent verification may fail. Crypto firms should also examine paths that create and display transaction data for approval, beyond just private keys. Limiting vendor access rights and account permissions, along with verifying transaction content through separate channels, are part of these measures.
AI-powered social engineering attacks also require vigilance. The FBI warned in September 2024 that North Korea-linked attackers targeted employees of crypto and DeFi firms using recruitment or investment opportunities. Attackers researched targets' careers, skills, and business interests before initiating contact. Anthropic disclosed in August 2025 that North Korean personnel used its AI model Claude to fabricate resumes, pass technical tests, and secure remote positions at U.S. tech companies.
While recruitment fraud and crypto theft are distinct activities, both cases underscore the importance of verifying external personnel identity and access management within security protocols. AI is also being used for defense, such as vulnerability detection and validation. Mozilla, in collaboration with Anthropic, identified 22 security vulnerabilities in Firefox in March 2026, 14 of which were classified as high severity.
Mozilla explained that reproducible test cases submitted by researchers aided engineers in verification and modification. Finding vulnerabilities and applying fixes to systems are separate steps requiring approval processes, patch schedules, and consideration of potential service disruptions. For exchanges, suspending accounts or restricting withdrawals can impact users and trading operations. The UK National Cyber Security Centre (NCSC) projected in May 2025 that AI tools could significantly enhance attackers' ability to exploit known vulnerabilities by 2027.
NCSC's Dave Chismon, Chief Technology Architect, advised defense teams in a September 21, 2026 post to consider budget, patch schedules, change approvals, and business disruption risks. He suggested expanding AI use for clear-scope tasks with easy recovery, such as information analysis and decision support.
The disclosed cases demonstrate that AI can accelerate work on both offense and defense. Crypto firms' response must go beyond identifying anomalies and validating risks to include restricting transaction permissions and blocking unauthorized fund movements.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: AI 공격, 자격 증명 탈취 작전 6시간 안에 수행