When I was at PwC, you needed IT approval to do just about anything on your laptop. It got stricter the longer I was there. Early on you could stay logged into a personal email account; then that went away. You could install Python libraries; by the time I left, that was blocked too. Since then I’ve worked with clients whose policies said the same things, but nothing actually stopped me. I could download whatever I wanted. It was all logged, but logging something isn’t the same as preventing it.
That gap between what a policy says and what a system enforces is how North Korean hackers took about $6.3 billion from Binance last week, the largest theft of any kind on record. Most of it moved in about 20 minutes. The way in was a job interview.
How it happened
Binance published its own report on Tuesday. The attack started in June, when a DevOps engineer on the wallet team searched online for a popular AI coding tool and ran a tampered installer. Binance’s security team caught it, replaced his laptop and reset his credentials. That’s the response you’d want, but the attackers had already learned his role and who he worked with.
In September they went after another engineer. Someone posing as a project director reached him through a legitimate job platform, likely because his profile listed Binance experience. He did a video interview with a fake HR person, then a second round with the “CEO” of a company that didn’t exist. At the end he was asked to download a coding assignment and get it running. He did that on his company-issued laptop. The file carried custom malware that antivirus didn’t catch.
On October 20, the attackers used his machine, along with an internal Binance AI tool, to take over two more engineers’ computers. On one of them was a deployment permission token. On October 29 they used that token to reach the service that signs wallet transactions and started withdrawals outside the normal approval process. Partners froze about $3 billion, and Binance says it’s covering the rest from company funds.
They knew this playbook
In September 2025, Binance’s founder and former CEO, Changpeng Zhao, posted a warning on X calling North Korean hackers “advanced, creative and patient.” He described them posing as employers, setting up interviews and sending sample code that installs malware. His advice was to train employees not to download files they don’t recognize. That’s good advice. The problem is that training is a policy, and a policy only works if every person follows it every time.
AI makes this harder. An eleven-country advisory in July confirmed these operatives use real-time deepfake video in interviews. A believable recruiter, CEO and company website can mostly be generated now. And AI shows up twice in Binance’s own report: the first way in was a fake AI tool, and later the attackers used Binance’s real internal AI tool to move between machines.
What this means for finance teams
Think about segregation of duties. Your policy says the person who sets up a vendor can’t also pay that vendor. An auditor doesn’t stop at the policy. They pull system access and check whether anyone can actually do both. If the ERP allows it, the policy isn’t preventing anything.
Most IT security policies work the same way. Don’t install software from a search. Don’t use your work laptop to interview elsewhere. Don’t run code from outside the company. Don’t leave credentials sitting on your machine. Binance almost certainly had all of these written down, and each one was broken. With a staff of a thousand people, expecting every one of them to follow every rule every time isn’t realistic. If a system lets someone do something, assume someone eventually will.
Two places I’d start:
Test your policies the way an auditor would. Can an employee install software on a company laptop without approval? Can they run a file from someone outside the company? If the answer is “they’re not supposed to,” ask IT what actually blocks it or forces an approval.
Map your approval path. Who can release a payment, change vendor bank details or approve a wire? What device do they use: a work laptop, a personal phone, a locked-down phone? Do they have that access all the time, or only when they need it? The token at Binance sat on a laptop between uses, and that’s what turned one compromised computer into a $6.3 billion loss. If your logins and tokens never expire, this is a good time to change that.
Key takeaways
- A policy only protects you if a system enforces it. Test it the way you’d test segregation of duties.
- People are the target, and AI has made impersonation cheap. Fake recruiters and coding assignments are a known attack route.
- If a system allows something, assume someone will eventually do it, whether that’s an employee, an attacker or an AI tool.
- Look for access that never expires. Long-lived credentials and tokens are what attackers look for.
- Your internal AI tools have access too. Know what they can reach.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: The $6.3B Job Interview
Leave a Reply