How hackers stole $6.3 billion from Binance in three stages

Binance lost about $6.3 billion in roughly 20 minutes. It is being called the largest crypto theft ever.

The 20 minutes are the part everyone remembers. We'd say the more useful part is the four months before it. Nobody guessed a password or found a magic flaw in the exchange's code. They went after people, one stage at a time.

Binance has published its own account, and it runs as a string of dated steps. We've grouped them into three stages: the tampered tool, the fake interview, and the October takeover and theft. If you read other write-ups, you may see the same events cut up differently.

Stage 1: a free AI tool that wasn't what it seemed

An engineer on the wallet team went looking for a popular AI coding tool and downloaded a tampered copy. Binance's account is that it came through a search engine result. Nothing about it looked like an attack. It looked like someone trying to get on with their day.

Binance caught it, replaced the device and changed the credentials. That wasn't the end of it. The engineer's role, and others on the wallet team, were by then exposed to the attacker.

Stage 2: a job interview with a company that didn't exist

Three months later, a different software engineer, looking for a new job, was approached on a job platform by what looked like a real company. He had two online interviews. One was a video call with its "CEO". The company did not exist.

Along the way he downloaded what he thought was a code repository, and it was malicious. That gave the attacker a foothold on a second machine.

Binance's first public line was that a software engineer's compromised laptop had been used for the withdrawals. The fuller account is much bigger than that, which is common in the first days after a breach.

Stage 3: take over the people, then take the money

On 20 October the attacker used the compromised computer together with one of Binance's own internal AI tools to take control of a wallet developer's system, and of a second wallet developer's system on the same team.

An internal tool, built to help staff, became a route to their colleagues and their logins. We've written about AI agents getting more access than anyone planned, and this is a close cousin. Whatever your tools can reach, a person who gets into them can reach too.

Then, on 29 October, the attacker used the second developer's computer, with the first developer's stolen credentials, to run a malicious script. Per Binance, that gave access to the wallet signing service, the part that approves money leaving.

There was a nasty twist. Binance said the attackers used an Ethereum feature called EIP-7702 to grant themselves permissions on the network itself. So pausing Binance's own withdrawals did not stop the transfers. The shutters were down and the money was still walking out.

What happened next:

• Stablecoin issuers Tether and Circle froze wallets, which locked roughly $3 billion in stablecoins. The Bitcoin could not be frozen.
• Binance said it would absorb the loss rather than its users, and by 3 November said 98% of the stolen tokens had been replenished. That figure is Binance's own, and we haven't seen it independently checked.
• The FBI has pointed at a North Korean group. North Korea has denied it. Binance has not named anyone itself.

An analyst at Elliptic told reporters it was "almost certainly the single largest-known theft of any kind."

What can a small business take from this?

You don't hold $6 billion in crypto. But look at the three stages again. Every one of them is something you can do something about.

• Only install from official sources. The June download was a lookalike of a real tool. Cyber Essentials' malware control asks you to limit software to approved, trusted sources for exactly this reason. Make it a rule: new tools go through one person, and nobody installs from a search result.
• Be careful with hiring. If you recruit, you're a target for the same trick in reverse. Never run code from a candidate, or a "company" you can't check, on a work laptop.
• After an incident, ask what that person could reach. June was "handled": new device, new passwords. The attacker still got through. Wiping the laptop isn't the same as closing the door.
• Keep money and admin access on separate, locked-down machines. The person who browses the web and takes interviews shouldn't be the same person, on the same laptop, who can approve payments or change who has access.
• Know what your AI tools can see. If a tool can reach customer data, shared drives or colleagues' accounts, treat getting into it as getting into all of those.

Quick answers

If you're not sure what your team can install, or who can reach what, our free 5-minute readiness check is a good start, or book a free 30-minute call and we'll talk it through.