Binance says hackers used an internal AI tool to steal $6.3 billion. In its timeline, that was one step in five months.

Binance’s incident report landed on November 3, and most headlines kept the same four words: internal AI tool. That’s accurate. It’s also a small slice of what the report describes, which is a campaign that started in June and ended in a four-minute withdrawal spree on October 29.

I couldn’t open Binance’s post on X directly, so everything below comes from outlets that read it: CNBC, The Verge, TechCrunch, The Block and Nikkei Asia. Where they quote the report, I say so.

How the break-in unfolded

Here’s the sequence as those outlets lay it out.

When	What happened	Reported by
June	A Wallet DevOps engineer searched for a popular AI coding tool and installed a tampered Homebrew package. Binance caught it, replaced the laptop and rotated credentials, but the attackers had learned who worked on wallet infrastructure.	CNBC, quoting Binance’s report
September	Another engineer got a job offer from a fake company, went through two interview rounds, and was handed a coding assignment that carried malware.	TechCrunch, The Verge
October 20	Using that laptop and an internal AI tool, the attackers took over two other engineers’ machines and found a container deployment permission on one of them.	The Block, Nikkei Asia
October 29	With the stolen credentials they deployed a container and ran a malicious script against Binance’s wallet-signing systems.	The Block

The money moved fast. Nikkei Asia reports 233.46 million XRP left five hot wallets in four minutes, and CNBC puts the total affected at $6.34 billion. According to The Block, Binance says it froze or intercepted about $2.85 billion, and CNBC reports it has replaced 98% of the affected assets from its own funds.

What the AI tool did, and what nobody has explained

On this part the outlets agree, and the report is thin. The Verge says the attackers used the internal tool to gain control of the first compromised engineer’s wallets and a third employee’s, and then pointed out that there’s “little detail” about what the tool was or why it helped them. CNBC quotes co-CEO Richard Teng saying the attackers “moved violently through our internal AI tools” on their way to the system that authorized wallet transactions. TechCrunch describes it more mundanely, as an internal tool capable of taking over the work computers of two other engineers.

My read: those three descriptions don’t match exactly, and a tool that can control engineers’ machines sounds like remote-access automation, which is a different thing from a chatbot someone prompted. I can’t tell which one it was, and none of the reporters say.

CryptoSlate adds a useful limit: Binance’s investigation doesn’t establish that an AI model planned or carried out the theft. The attackers used an internal AI tool as one piece of a campaign built on stolen credentials, tampered developer software and social engineering.

The part that wasn’t new

The entry methods are old. TechCrunch notes that researchers have warned for years about North Korean operators posing as recruiters and investors, and the FBI has linked this attack to a group it calls TraderTraitor. North Korea’s Kim Yo Jong rejected that on Monday, according to The Block. Teng, per CNBC, didn’t name North Korea himself and acknowledged the FBI’s finding.

The Verge also reports why the withdrawals were hard to stop. Binance’s hot wallet used multi-party computation, so “there was no single private key that could be used to immediately revoke the malicious EIP-7702 authorizations.” Binance gives that as the reason it couldn’t cut off the transfers that followed the first unauthorized ones.

My read

My read: the lazy version of this story is “AI got hacked, AI is dangerous”. The report supports something narrower and, for companies, more uncomfortable. An internal tool with enough reach to control engineers’ machines is a lateral-movement path, and it doesn’t matter much whether it has “AI” in its name. I covered the other direction in the Cairn study of AI agents hacking company sites, where the agents did the attacking. Here the attackers appear to have borrowed access that already existed.

My read: Binance’s own rebuild list is the tell. It says it rebuilt the wallet system and the internal AI tool, replaced devices and retired tokens, and plans to put more money into AI-based threat detection. Fixing a tool by rebuilding it suggests it was a real weak point, but Binance hasn’t said what was wrong with it.

Two things to watch. Binance says an independent firm is reviewing the incident, and that review is where the AI tool should get a proper description. And if it’s published, check whether it names the tool’s permissions, because that’s the detail other companies can actually learn from.

Note on sources: the timeline and figures come from CNBC, The Verge, TechCrunch, The Block and Nikkei Asia, each attributed in the text. The Verge, CNBC and The Block quote or summarize Binance’s report, which I read only through them.

Sources

1. CNBC: Binance co-CEO Richard Teng says hackers used fake job interviews, AI tool to steal $6.3 billion in crypto
2. The Verge: Binance says hackers used fake job interviews and an internal AI tool to pull off the $6.3 billion heist
3. TechCrunch: Researchers warned of North Korean hacking tactics years before record $6.3B Binance heist
4. The Block: Binance says hackers used fake job interview, internal AI tool in $6.3 billion breach
5. Nikkei Asia: Binance hackers used fake job interview, tampered software in $6.3bn attack
6. CryptoSlate: Binance relied on AI to fight fraud before hackers exploited internal tool in $6 billion theft