Binance Plans AI Controls After Oct. 29 Security Incident

Binance is reviewing an Oct. 29 security incident and plans to place its internal AI tools and development environments under critical-infrastructure standards, Chief Security Officer Jimmy Su said. The exchange is also working with AI and security companies to establish a research fund for security threats involving blockchain and AI, according to Su.

The proposal came without a disclosed fund size, named participants or launch date. Su’s account also did not quantify losses, identify the chain or token involved, or specify how many transaction authorizations attackers used.

What did attackers exploit?

Su said attackers used “transaction authorizations recorded on the blockchain,” outside Binance’s withdrawal system. That separation defeated a routine response: pausing withdrawals would not stop the outflow because the authorizations did not depend on the exchange processing withdrawals.

Under Su’s account, Binance had to submit a separate on-chain transaction to cancel those authorizations. The response therefore required intervention in the authorization layer rather than merely freezing the exchange’s withdrawal interface.

How did wallet design affect the response?

Binance’s hot wallets use multi-party computation, Su said. Because no single private key could revoke permissions immediately, the incident team had to develop a revocation process while responding to the attack.

The resulting process took several hours to bring the situation under control. That delay turned multi-party control into an operational constraint during an active outflow, requiring a coordinated revocation process rather than a unilateral key action.

What role did malware play?

Su also said the attack used “zero-day malware that third-party endpoint protection software did not detect.” He did not describe the initial access path, the malware’s persistence or the number of affected endpoints in the account he provided.

That gap matters at the boundary between endpoint defenses and on-chain controls: the withdrawal pause could not address authorizations already recorded, while the endpoint product failed to identify the malware used in the incident.

What changes for AI and development tools?

Binance plans to manage internal AI tools and development environments under critical-infrastructure standards. Su did not name a regulatory regime, certification or completion deadline, so the concrete scope remains an internal control commitment rather than a specified compliance program.

For the exchange, the stated change links two parts of its security posture: governance of AI tools and development environments, and controls over transactions that can move assets outside the normal withdrawal interface. The first addresses the environment in which security processes are built; the second addresses how Binance responds when an authorization survives a withdrawal freeze.

Su’s account describes a Binance plan; it does not identify a regulator, external mandate or enforcement proceeding. That makes the standards an internal control commitment, not a disclosed compliance order.

What is the proposed blockchain-AI fund meant to address?

The planned research fund would focus on threats involving blockchain and AI, according to Su. Binance is working with AI and security companies to establish it, but the account did not disclose a target size, participant list, governance structure, specific research agenda or funding timetable.

The available details establish a proposal, not a disclosed funding program. Its stated remit reaches beyond the Oct. 29 response by pairing AI-focused research with security work tied to blockchain systems, although Su did not name participants or assign deliverables.

Su’s description places two failure modes at that intersection: an authorization-based attack can bypass controls around withdrawals, while an undetected zero-day can defeat endpoint monitoring. Su did not say whether the fund would examine Binance’s incident directly.

What remains undisclosed?

Su did not provide the incident’s financial size, the affected chain or token, the number of transaction authorizations used by attackers, or a full account of the malware’s entry point. The account also gives no deadline for critical-infrastructure standards to cover every internal AI tool and development environment, or for the research fund to begin operating.

That leaves three practical tests for the next disclosure: whether the revocation process can operate faster than the several-hour response, whether critical-infrastructure standards establish measurable controls for internal AI tools and development environments, and whether the research fund arrives with defined funders, mandates and milestones.

Absent those details, the most grounded measure of change is operational: Su says Binance developed a revocation process, but he did not say the exchange had shortened the multi-hour containment time or prevented recurrence.

The immediate forward test is whether Binance can institutionalize the incident-derived revocation process before another authorization-based attack; Su set no deadline for that transition or for the research fund.