The Battle for AI Power

Part 2 · Analysis

The $6.3 Billion Heist That Began With a Job Interview

A fake recruitment process, compromised development tools and an internal AI system were reportedly part of the attack on Binance. The wider lesson reaches far beyond cryptocurrency.

Harbour Creative TeamHarbour Creative Team

8 min read

Keep this story in your reading list
The $6.3 Billion Heist That Began With a Job Interview
Share article
0%
Ways to read

The Battle for AI Power — Part Two

A software engineer receives an invitation to interview for a new job. The conversations seem legitimate. The prospective employer appears professional. Then comes a technical assessment, the kind developers complete every day.

The engineer downloads the assignment.

Weeks later, one of the world's largest cryptocurrency exchanges loses more than $6 billion.

That is the sequence emerging from Binance's preliminary investigation into the 29 October 2026 attack, which the FBI attributed to North Korea's TraderTraitor operation.

The number is extraordinary. But the way the intrusion reportedly unfolded may be more revealing than the record itself.

There was no need to defeat an entire organisation in a single attempt. According to Binance, the attackers spent months working through familiar workplace activities: developer downloads, recruitment, employee devices and internal services.

The eventual theft raises an uncomfortable question. What happens when technology companies become so complex that the systems meant to make them efficient also provide new routes for attackers?

The interview that wasn't an interview

The operation reportedly began taking shape months before the theft.

In June, a wallet infrastructure engineer downloaded a tampered software installer while trying to set up an AI coding tool. Binance says its security team detected the anomaly, replaced the device and rotated credentials.

Yet, according to the exchange's account, information about the engineer's role and colleagues was exposed.

Then, in September, a second engineer was approached through a legitimate job platform by someone pretending to represent another company.

The conversations moved through what appeared to be ordinary recruitment stages. The supposed interviewers held a video call and shared social-media material to establish credibility.

Finally, the engineer was asked to download code for a technical assessment.

The repository was malicious. Binance reports that its contents executed harmful code during setup, helped the attackers establish access and evaded the endpoint protection then in place.

The recruitment approach mattered because it exploited something that technology workers routinely regard as normal. A potential employer asking a developer to review or run code would not, by itself, seem extraordinary.

That ordinary appearance was central to the deception.

A software engineer working at a computer workstation

Developer recruitment and software installation were among the routes described in Binance's preliminary account. Illustrative photograph.

The AI tool inside the attack

One of the most striking details in Binance's report concerns an internal AI tool.

The exchange says that on 20 October, the attackers used access through the compromised second engineer's computer and an internal AI tool to gain control over the devices of two more employees.

They subsequently discovered a container-deployment permission token on one of those devices.

This is not evidence that AI independently planned or executed the theft. It is evidence, according to Binance's preliminary findings, that a company AI tool became one part of the pathway through interconnected internal systems.

That distinction is important.

AI assistants with narrow permissions present one sort of risk. Internal tools connected to company data, developer environments and privileged operations can present another.

The problem is not necessarily the intelligence of the software. It is the authority available to anyone who gains access to the software and the surrounding accounts.

A system designed to save employees time may also expand the consequences of a compromised account if its connections and permissions are not sufficiently contained.

For an industry rushing to put AI inside everyday business operations, that should be a serious point of examination.

Nearly four hours, billions in transfers

On 29 October, unauthorised transfers began moving through several Binance wallets and blockchain networks.

According to Binance's 3 November incident report, attackers gained unauthorised access to an internal wallet-signing service, enabling transactions outside the normal withdrawal process.

Binance says the incident affected approximately US$6.34 billion in digital assets. Its account describes transfers involving XRP, Solana, Bitcoin, Ethereum and other networks.

The exchange suspended withdrawals and coordinated with other companies to freeze affected assets. Yet stopping normal withdrawals did not immediately end every unauthorised transaction.

Part of the problem concerned Ethereum's EIP-7702 authorisation mechanism, which allows an account to delegate certain execution rights to smart-contract code without exposing its private key.

Binance says the attackers had created malicious authorisations capable of operating independently of its ordinary withdrawal infrastructure.

These permissions needed to be revoked through additional on-chain actions. The exchange says its multi-party computing arrangement added complexity to that response.

The final unauthorised transfer was confirmed at 23:56 UTC, almost four hours after the initial wallet movements.

As of its 3 November report, Binance said around $2.85 billion had been frozen or intercepted. It had also set aside sufficient resources to cover affected customer assets and said 98% of the impacted token assets had been replenished as of 31 October.

Those figures should not be confused. Frozen assets, recovered assets, funds reserved to reimburse customers and assets actually stolen describe different parts of the incident.

Portrait of Binance leadership

Binance said it would absorb the financial impact on its customers and commissioned further security review.

This was not the first warning

The Binance incident has a major historical precedent.

In February 2025, attackers stole approximately $1.5 billion from cryptocurrency exchange Bybit. The FBI attributed that theft to North Korean TraderTraitor actors.

The parallels are worth investigating, but the technical routes were not identical. What they share is the extraordinary scale of losses associated with attacks on trusted systems and transaction authorisation.

The wider lesson is older still.

In 2020, the SolarWinds compromise showed how attackers could use trusted software distribution channels to reach organisations that believed they were installing legitimate updates. The operation affected government agencies and private-sector systems.

There is no evidence that the SolarWinds operation and the Binance theft were connected.

The comparison is useful for a different reason. It exposes a recurring problem: organisations depend on software, accounts and other systems whose apparent legitimacy gives them authority.

Attackers often seek to inherit that authority instead of breaking every security barrier directly.

Binance's report describes an especially stark version of that pattern. A job interview led to malicious code; compromised access reportedly led through internal tooling; and permissions eventually reached systems able to authorise enormous transfers.

The state behind the allegation

On 31 October, the FBI publicly attributed the Binance theft to North Korea's TraderTraitor activity.

That attribution moves the incident beyond the usual story of cybercriminals stealing from a financial company.

Cryptocurrency can be transferred across multiple networks and intermediaries. The FBI said it had observed conversion of stolen assets into Ether and anticipated further laundering through decentralised finance and swap services.

The agency called on exchanges, infrastructure operators and analytics firms to help block transactions associated with the theft.

The claims are those of the FBI; a public attribution is not the same as a complete court-tested account of every participant in the operation.

Nevertheless, the response shows the extent to which private financial infrastructure and state security concerns now overlap.

A failure inside one company can become a matter for international law enforcement, blockchain infrastructure providers, financial services companies and governments.

View of Pyongyang, North Korea

The FBI attributed the Binance attack to North Korean actors operating under its TraderTraitor designation.

Ten days later, Nvidia

On 8 November, Nvidia announced an independent investigation into reports that CEO Jensen Huang had been targeted by a foreign intelligence effort.

The company said it had no reason to believe its commercial secrets or US national security had been compromised.

The Nvidia controversy and the Binance theft involve different organisations and different reported circumstances. There is no established evidence that the incidents were connected or coordinated.

Yet the proximity of the news invites a broader question about what powerful technology companies have in common.

At Binance, investigators have been examining how attackers moved from employee systems into privileged financial infrastructure.

At Nvidia, the issues concern allegations about an influential executive, foreign intelligence targeting and access to semiconductor technology.

Both cases put the spotlight on trust: who possesses it, what access it provides and what happens when an organisation must investigate whether that trust has been abused.

Neither incident, by itself, proves a systemic failure across the entire technology industry. Together, however, they make a compelling reason to examine how influence and access are concentrated.

What should concern every company building with AI

Technology firms have spent years investing in faster computing, more automated operations and increasingly capable internal tools.

Those advances can create real advantages.

They can also create systems in which a single compromised account, software dependency or device has consequences well beyond its original purpose.

The central concern exposed by Binance's preliminary investigation is not merely that an engineer downloaded malicious software. It is the alleged progression from that compromise into infrastructure with the authority to move enormous sums.

It is worth asking four simple questions of every expanding technology operation.

Who can access its most important systems? What can those people or tools do once inside? How quickly can their authority be revoked? And who is independently checking that the controls still work?

Those are questions of governance as much as cybersecurity.

The Binance heist was a financial crime on an extraordinary scale. Its broader significance lies in the way ordinary-looking workplace interactions reportedly became a route to extraordinary power.

The next great contest for technological influence will not be decided by processing speed or market value alone.

It will also depend on who can protect the people, permissions and infrastructure on which that power rests.


This is Part Two of The Battle for AI Power, a Harbour Creative Blogs series examining the companies, governments and individuals competing for control of critical technology infrastructure.

Reporting and sources

About this article

This article provides analysis and interpretation of available information. Sources and supporting material, where applicable, are linked within the article.

Found an error? Contact the editorial desk. See our editorial standards and corrections policy.

From the Harbour Creative Blogs archive

Keep exploring this story

Related reporting, people and places from our published coverage. These are editorial connections, not a popularity ranking.