Binance relied on AI to fight fraud before hackers exploited internal tool in $6 billion theft

Hackers exploited Binance's internal artificial intelligence tools to compromise engineers before stealing $6.34 billion in cryptocurrency, the exchange disclosed.

In its Nov. 3 security report, Binance said attackers used a compromised employee computer and an internal AI tool on Oct. 20 to take control of two engineers responsible for its wallet infrastructure.

The attackers subsequently obtained access to a container deployment permission token, which helped them penetrate the exchange's wallet-signing operations on Oct. 29 and initiate unauthorized transactions across several blockchain networks.

The disclosure puts AI at the center of one of the largest cryptocurrency thefts on record, adding to concerns about how increasingly connected AI systems can introduce new vulnerabilities into corporate infrastructure.

Binance's investigation does not establish that an AI model autonomously planned or executed the theft. Rather, the attackers exploited an internal AI tool as part of a broader campaign involving stolen credentials, compromised developer software, and social engineering.

In June, a Binance engineer downloaded a tampered software installer while attempting to install a popular AI coding tool. Binance detected the suspicious activity, replaced the device and rotated associated credentials, although information about the employee and colleagues had already been exposed.

In September, another engineer fell victim to a fake recruitment operation involving staged job interviews and a malicious coding assignment. The malware compromised the employee's computer and gave the attackers access, which they later used to move deeper into Binance's systems.

The FBI has attributed the broader attack to North Korea, whose hacking operations have increasingly targeted software developers and cryptocurrency businesses.

AI becomes a growing weapon in cyberattacks

The Binance incident follows months of warnings from cybersecurity researchers that artificial intelligence is expanding both hackers' capabilities and the systems available for exploitation.

In February, Google's Mandiant reported that North Korean hackers targeting the cryptocurrency industry had incorporated AI-assisted techniques into their operations.

The researchers documented an intrusion involving a fake video conference, malicious software, and a suspected AI-generated video impersonating a corporate executive. They could not independently confirm the deepfake in that particular incident but identified the group's use of AI tools for reconnaissance and social engineering.

By September, Google's Threat Intelligence Group reported that attackers were moving beyond using AI to generate malicious content toward automated operations that could execute multiple stages of an intrusion.

In one case, hackers compromised a cloud resource and organized an AI-assisted mass credential-harvesting campaign in less than six hours. Google also identified attackers attempting to manipulate AI coding assistants and security scanners during software supply-chain compromises.

The findings closely resemble one aspect of the Binance breach: developers and the AI-powered tools they depend on are becoming targets attackers use to reach sensitive systems.

Separately, AI developer Anthropic disclosed in September that criminals and suspected state-sponsored operators had used its Claude models to conduct reconnaissance, develop exploitation tools, and automate data theft.

Some operations involved AI agents carrying out multiple stages of an attack with limited human supervision, reducing the technical expertise and manpower traditionally required.

Binance turns to AI companies to strengthen security

The exchange is now seeking support from the same industry whose technologies were featured in the attack.

Binance said it is working to establish a security research fund and develop targeted initiatives with leading AI companies, blockchain industry participants and cybersecurity firms.

The proposed collaboration will focus on strengthening AI security in blockchain environments, although the exchange has not disclosed the fund's proposed size, participating companies or launch date.

It also plans to expand spending on AI-enabled threat detection, endpoint protection, developer-environment security and behavioral anomaly monitoring.

The exchange said it has already rebuilt the affected internal AI tool and wallet infrastructure, replaced compromised employee devices and strengthened access controls.

Binance has also commissioned an independent cybersecurity review to validate its investigation and identify additional safeguards.

The company said approximately $2.85 billion of the $6.34 billion in affected digital assets has been frozen or intercepted, while 98% of impacted token assets had been replenished as of Oct. 31.

The independent review has yet to establish the full extent of the weaknesses that allowed attackers to exploit the exchange's internal AI environment. Its findings will inform the security changes Binance plans to implement as it develops the proposed industry fund.