Binance’s $6.3B Heist Exposes AI Governance Gaps [2026] - shattered.io

Binance disclosed the security breach publicly, with reports indicating the attackers walked away with roughly $6.3 billion in virtual assets. The exchange flagged a potential security issue affecting approximately this amount, with other sources putting the figure at $6.34 billion. Most of the funds moved within roughly 20 minutes, suggesting the attackers had everything staged before executing the withdrawal requests.

The exchange stated that a compromised DevOps engineer's laptop may have allowed an attacker to reach a service container and submit unauthorized withdrawal transactions. Binance emphasized it found no evidence of any wallet private key leak, meaning the attackers didn't need to steal keys directly but instead manipulated the signing and approval workflow.

Withdrawals were frozen while the company conducted checks before being reopened. By the time Richard Teng, Binance's co-CEO, went public with the intrusion details, the focus had shifted from the amount stolen to how the breach occurred—a question that will likely shape security budgets in 2027.

Attack Chain

The campaign combined fake recruiters or job interviews, malicious downloads, compromised employee computers, and an internal AI tool. SecurityWeek framed it as a rigged coding assignment and an abused AI tool opening the path, while The Verge reported hackers used fake job interviews and an AI tool to execute the theft.

The entry point was social engineering, with candidates receiving coding exercises that contained malicious code. Teng described it as a "more complex attack process" rather than a single "one-click" action, suggesting a campaign built over time rather than a drive-by exploit.

AI Tool Exploitation

According to CSO Online, attackers used Binance's internal AI tool to reach additional developer systems, obtain an access token, and deploy a container on Binance infrastructure. The AI tool served as a bridge from the compromised machine to more critical systems.

The AI assistant likely had broad permissions—reading repositories, calling internal APIs, accessing tickets and logs, and potentially holding tokens. When controlled by an attacker through a compromised laptop, it became a lateral-movement service with extensive access to the network.

Binance reported rebuilding the affected wallet system and the AI tool, replacing devices, and retiring compromised credentials. Rebuilding rather than just patching the AI tool suggests the company now views such assistants as part of the trusted boundary that can be compromised.

Security Failures

The incident revealed multiple security failures beyond the AI exploitation. Binance detected a first malware installation in June, but endpoint security tools missed a subsequent attack where a malicious file replaced the pyenv executable—a tool Python developers use to manage interpreter versions.

The most technical finding concerns Ethereum's EIP-7702 delegation feature, which the attackers used to bypass Binance's normal approval controls. This capability allows externally owned accounts to delegate execution to smart-contract code, potentially slipping past approval systems designed only for traditional transfers.

The incident highlights a classic gap between protocol evolution and internal policy. As Ethereum added new capabilities, exchanges needed to update their approval logic to maintain security guarantees. The same lesson appeared when LayerZero moved away from single-verifier configurations after a $292 million loss.