Binance has disclosed how attackers stole $634 million in crypto assets, using a compromised employee computer and the exchange’s internal AI tools to gain control of two wallet infrastructure engineers. The attackers later accessed wallet-signing operations and made unauthorized cross-chain transactions. Binance says about $285 million in affected assets has been frozen or intercepted, while the incident highlights the links between AI tools, development environments and employee credentials as potential points of entry.
Compromised devices led to wallet-signing access
Binance said the attackers took control of the two engineers on October 20 using a compromised employee computer and internal AI tools. They later obtained container deployment access tokens and, on October 29, entered the exchange’s wallet-signing workflow to initiate unauthorized transactions across multiple blockchain networks.
The information disclosed so far does not indicate that an AI model independently planned or carried out the theft. Instead, the attack involved stolen credentials, compromised development software and social engineering, with an internal AI tool forming one part of the chain.
The investigation also traced earlier intrusions targeting employees. In June, a Binance engineer downloaded a tampered installer while installing a popular AI coding tool. Binance detected unusual activity, replaced the device and rotated the relevant credentials, but information about the employee and colleagues had already been exposed.
In September, another engineer was targeted through a fake recruitment process. Attackers arranged fictitious interviews and infected the engineer’s computer through a malicious coding assignment. Binance said the device was then used to gain further access to internal systems. The FBI attributed the broader campaign to North Korea.
Security teams warn of AI-assisted attacks on developers
The Binance incident comes amid continued warnings from cybersecurity researchers that AI is expanding attackers’ capabilities while creating more potential entry points into corporate systems. In February, Google’s security team Mandiant disclosed that North Korean hackers targeting the crypto industry had used AI-assisted techniques.
Researchers have documented an intrusion involving a fake video meeting and malware, and suspect attackers used an AI-generated video to impersonate a corporate executive. They could not independently confirm whether the video was a deepfake, but found that the group used AI tools for reconnaissance and social engineering.
Google’s threat intelligence team said in September that some attackers were moving beyond using AI to generate malicious content and were applying it to automate stages of intrusions. In one case, hackers organized a large-scale, AI-assisted credential theft operation less than six hours after compromising cloud resources. Google also found attempts to manipulate AI coding assistants and security scanning tools as part of software supply-chain attacks.
Anthropic said that month that criminals and suspected state-backed actors had used its Claude model for reconnaissance, developing attack tools and automating data theft. Some operations involved AI agents carrying out multiple stages with limited human oversight. The Binance case shows how developers and the AI tools they rely on can also provide a route into sensitive infrastructure.
Binance rebuilds systems and plans industry collaboration
Binance said it has rebuilt the affected internal AI tools and wallet infrastructure, replaced compromised employee devices and tightened access controls. The exchange has also commissioned an independent cybersecurity review to verify the investigation’s findings and identify further safeguards.
The company plans to increase investment in AI-assisted threat detection, endpoint protection, development environment security and behavioral monitoring. It also intends to establish a security research fund and pursue targeted collaboration with AI companies, blockchain industry participants and cybersecurity firms. Binance has not disclosed the fund’s size, participants or launch date.
Of the $634 million in affected digital assets, about $285 million has been frozen or intercepted, Binance said. As of October 31, 98% of the affected token assets had been replenished. The independent review has not yet established all the vulnerabilities exposed through the attackers’ use of Binance’s internal AI environment; its findings are expected to inform further security measures and plans for the proposed industry fund.