*Editor’s note: This analysis reflects reporting available as of November 4, 2026, including Binance's initial incident report and the FBI's attribution. The investigation remains ongoing, and additional findings may change our understanding of the attack.
What to Expect in This Blog
Attackers don’t always need to steal the keys to the kingdom. Sometimes they only need access to a system that already has permission to open the doors.
On October 29, 2026, the Binance hack became the largest recorded cryptocurrency theft, with approximately $6.34 billion in digital assets affected. Most of the funds reportedly moved in about 20 minutes, but the attack had been months in the making.
Binance's November 3 investigation report describes a sophisticated campaign involving malicious developer software, fake job interviews, compromised engineering workstations, and stolen credentials. Attackers ultimately gained unauthorized access to the exchange's wallet-signing infrastructure, allowing them to initiate transactions outside the normal withdrawal process.
Perhaps the most revealing detail is what Binance has not found: evidence that wallet private keys were stolen. For security leaders, the lesson reaches far beyond crypto. The greatest risk is not always who holds the secret. It is who, or what, has the authority to use it.

The Binance hack illustrates why least privilege matters even when cryptographic keys remain protected. Attackers who compromise trusted systems may still be able to execute critical operations if permissions and authorization controls allow them.
Independent estimates vary with valuation timing and methodology. TRM Labs estimated roughly $6.36 billion moved across multiple blockchains, while the FBI attributed the attack to North Korea's TraderTraitor operation.
Binance suspended withdrawals, revoked affected credentials and authorizations, and worked with blockchain partners to freeze assets. By November 3, approximately $2.85 billion had been frozen or intercepted.
Binance began restoring withdrawals in phases on November 1 and says independent experts will review the incident. A full technical post-mortem is still to come.
The uncomfortable question remains: how could attackers reach systems able to move billions in minutes without extracting wallet private keys?
How Compromised DevOps Workstations Exposed Critical Systems
A DevOps engineer's laptop is not an ordinary endpoint. DevOps teams build, deploy, configure, and troubleshoot an organization's most critical systems. Their workstations can reach cloud environments, Kubernetes clusters, container registries, CI/CD pipelines, production infrastructure, secrets managers, service accounts, and admin interfaces.
Compromise a standard employee laptop, and you may get email and business apps. Compromise a DevOps workstation, and you may gain access to critical production systems, depending on how permissions and credentials are managed. An attacker may only need to hijack an authenticated session, misuse a token, or abuse a trusted workflow.
How the Binance Attack Unfolded: June to October 2026
On November 3, Binance published its most detailed incident report so far, and it reinforces the central argument of this article: there was no single point of failure. Co-CEO Richard Teng told CNBC the attackers “spent months building their way in.”
According to Binance's report, the campaign unfolded in three stages:
- June: A poisoned developer tool. A Wallet DevOps engineer downloaded a tampered Homebrew installation while searching online for a popular AI coding tool. Binance detected it, replaced the device, and rotated credentials, but by then the attackers had learned the engineer's role and who his colleagues on the wallet infrastructure team were.
- September: A fake job interview. A second engineer, whose profile on a legitimate job platform mentioned his Binance experience, was approached by someone posing as a project director. After a video interview with a fake HR representative and a second round with a fake “CEO” on a chat app, he was asked to download a coding “assignment” onto his work laptop. It contained malware designed to evade security tools.
- October: From one engineer to the wallets. On October 20, using that compromised laptop and an internal AI tool, the attackers gained control of two more engineers' machines. One of them held a container deployment permission token. On October 29, the attackers used stolen credentials to deploy a malicious container and reach wallet-signing operations. They also exploited EIP-7702 to establish malicious execution permissions over a hot wallet, moving funds outside Binance's normal withdrawal process.
Identity is the connective tissue of the modern business, and this attack moved along it one link at a time. The attackers did not need to defeat cryptography itself. They targeted the identities, permissions, and trusted workflows surrounding it.
Rotating credentials is not the same as containing exposure. Binance rotated credentials after the June compromise, but the attackers had already mapped who mattered. When someone in a privileged role is compromised, treat their team and the access paths around them as potentially exposed too.

They Didn’t Steal the Keys. They Asked the Vault Nicely.
In crypto infrastructure, private keys sign transactions. Organizations protect them with hardware security modules, dedicated signing infrastructure, and sophisticated custody architectures. Those protections matter, but a protected key does not guarantee that every transaction signed with it is legitimate.
Imagine a vault with an exceptionally strong lock, its key sealed inside a tamper-resistant device. Now imagine an internal service with permission to ask for the vault to be opened. If an attacker compromises that service, and its requests are trusted without independent authorization, the vault opens anyway. The key doesn't need to be stolen. The authorization process has been abused.
How EIP-7702 Delegation Enabled Unauthorized Transactions
Binance has now confirmed that attackers gained unauthorized access to its internal wallet-signing service and abused Ethereum's EIP-7702 authorization mechanism. EIP-7702 allows a wallet to delegate certain execution rights to smart contract code without exposing its private key. In this case, the attackers used the signing service to establish malicious delegations that allowed an attacker-controlled contract to execute unauthorized transfers.
Other affected blockchains used ordinary signed transfers, reinforcing the broader concern about unauthorized access to Binance's signing infrastructure.
According to TRM Labs' on-chain analysis, the attackers prepared 50 pre-signed EIP-7702 authorizations for different account nonce values. Once a valid authorization was established, the attacker-controlled contract could continue moving assets independently of Binance's normal withdrawal infrastructure. Suspending withdrawals and shutting down hot wallet services did not automatically revoke that delegation.
Why Stopping the Transactions Was So Difficult
Containment was further complicated by Binance's Multi-Party Computation (MPC) signing architecture. Because no single private key could be used to revoke the malicious EIP-7702 authorizations immediately, the team had to develop and execute a separate revocation procedure. The final unauthorized transfer settled at 23:56 UTC, nearly four hours after the first reported movements.
The private keys may have remained protected, but the authority to use the signing infrastructure had been compromised. That's the distinction security leaders need to understand.
How Least Privilege Limits the Damage
Standing privileges are permissions that stay available even when nobody needs them. A DevOps engineer may need admin access to a production Kubernetes cluster to troubleshoot, but should that access be available around the clock? A CI/CD pipeline may need to deploy an application, but should it also modify security controls, retrieve production secrets, or initiate financial transactions? A service account may need to talk to a signing service, but should it request unlimited operations across every wallet and environment?
Too often, permissions are granted to fix an immediate problem and never revisited. Temporary access becomes permanent. Service accounts inherit broad roles. Admin tokens are reused across environments.
Five Ways to Enforce Least Privilege
The principle is simple: every human or machine identity, application, service, or AI agent gets only the permissions its task requires, for only as long as it requires them.
Applying it well takes more than removing local admin rights. Modern least privilege must be dynamic, contextual, and enforced across the whole identity ecosystem.
- Eliminate unnecessary standing privileges. Grant just-in-time (JIT) privileged access for a specific purpose. Deployment rights should be scoped to one task, bound to a managed device or workload, and revoked automatically when no longer needed. An engineer restarting a service should not also gain authority over unrelated workloads or transaction systems.
- Separate development, deployment, and production authority. A compromised developer laptop should not reach production transaction authorization. A deployment pipeline should not hold unrestricted authority over everything it can touch. Separate identities, roles, and trust relationships across environments.
- Apply least privilege to machine identities. Service accounts, API keys, workload identities, and automation tokens often run continuously with little oversight, and some hold more power than the people managing them. Discover them, map their relationships, restrict their permissions, and monitor them. Replace persistent secrets with short-lived, workload-bound credentials wherever possible.
- Enforce authorization at the point of action. Authentication says who you are; authorization says what you may do. A properly authenticated service should not be trusted to perform every operation it requests. High-impact actions need contextual checks on the operation, target, value, and risk, with independent approval and separation of duties for the most sensitive.
- Prevent privilege escalation and lateral movement. A workstation compromise should stay a workstation compromise. Restrict access to cloud roles, service accounts, secrets, production management interfaces, and privileged execution environments so one compromised identity cannot become an organization-wide incident.
Zero Trust Must Extend Beyond the Login
Some Zero Trust implementations still focus heavily on initial authentication: MFA, device posture, conditional access. Those controls are valuable, but they fall short if a trusted identity can then execute dangerous operations without further scrutiny. An attacker on a compromised endpoint can ride an existing session or a trusted application identity, and from that point the question changes. It is no longer whether the identity authenticated. It is whether this specific action should be allowed.
A service that normally handles a modest volume of withdrawals should not be able to move billions without triggering additional controls. A deployment identity should not suddenly reconfigure privileged infrastructure. A developer workstation should not reach transaction-signing services just because it authenticated to an internal network. Zero Trust should not end when access is granted. It must continue through every privileged action.
ITDR: Detecting When Trusted Access is Abused
Traditional monitoring looks for malware, suspicious traffic, and known indicators of compromise. But when an attacker works through a legitimate identity, individual requests may appear legitimate: the service account is recognized, the token is accepted, the transaction is properly signed. Only the behavior is wrong.
Identity Threat Detection and Response (ITDR) helps close that gap by spotting abnormal identity behavior and privilege abuse. In a scenario like the reported Binance path, useful signals include unusual access to a production container, unexpected token use, abnormal privilege elevation, shifts in workload identity behavior, and sudden spikes in sensitive transaction requests.
Detection alone is not enough. Most of the stolen funds reportedly moved in about 20 minutes, faster than many teams can detect, investigate, and contain a complex incident. Organizations need preventive authorization controls and transaction limits, backed by automated containment: revoking sessions, disabling service credentials, suspending privileged operations, and holding high-risk actions for independent approval. These are layered safeguards, not a guarantee against any single exploit.
Binance reported that the attackers used sophisticated evasion techniques to avoid detection and remove traces of their activity. When attackers can erase evidence from compromised systems, logging for privileged sessions, machine identities, and authorization operations must be independent, tamper-resistant, and beyond their reach.
Detection must operate alongside enforcement, not replace it.
Securing Privileged Access in DevOps Workflows
DevOps runs on speed and automation. Controls that add heavy friction push teams toward workarounds, shared credentials, and unmanaged access. The answer is not to slow DevOps down. It is to make secure access the easiest path.
Coding tests and developer tools are untrusted code. Interview assignments, package installs, and tools found through search belong in isolated environments, never on a machine that holds production credentials.
Engineers in privileged roles are targets. Binance says it blocked 57 suspected North Korea-linked attempts this year. Awareness training for these teams should cover fake recruiters and coding tests, not just phishing emails.
At the same time, organizations need to build privileged access controls into developer workflows, CI/CD pipelines, and cloud infrastructure. Developers should have temporary access to the specific production resources they need, pipelines should use narrowly scoped credentials, and production services should have clearly defined permissions. Sensitive actions should remain subject to policy whether the request comes from a human or a machine.
None of these controls would stop every social engineering attempt. That isn't the goal. The goal is that when one engineer is fooled, and eventually one will be, the credentials within reach are too limited, too short-lived, and too well-watched to become a $6.3 billion problem.
Why Privileged Access Management Must Go Beyond Credentials
For years, Privileged Access Management (PAM) has focused on securing admin credentials, managing privileged sessions, and controlling access to critical infrastructure. Those capabilities remain fundamental. But privileged access now lives across human identities, machine identities, cloud workloads, containers, APIs, automation platforms, and AI-driven systems.
A service account capable of authorizing billions in transactions may represent a greater immediate business risk than a traditional administrator account. A deployment pipeline may be able to change infrastructure across several environments. An AI agent may be allowed to act on its own.
Privilege is not defined by a job title or account classification. It is defined by the authority an identity holds and the consequences of abusing it.
That calls for a broader approach to privileged access security that brings together discovery, privileged access enforcement, authorization governance, monitoring, and automated response. Organizations need to know who and what has access, why, how it is used, and whether it is still appropriate. Above all, they need the ability to stop a dangerous action before it becomes a catastrophic one.
Five Questions Every CISO Should Ask After Binance
This is not only a crypto problem. Ask these questions about every environment where trusted identities can trigger high-impact business operations.

Answer these through testing, not policy documents. Simulate a compromised developer workstation, map the permissions its identities hold, and measure how far an attacker gets before hitting an effective control. Then test whether suspicious privileged operations are stopped fast enough to prevent material impact. The goal is not to prove a workstation can never be compromised. It is to prove you can contain the consequences when one is.
What the Binance Hack Means for AI Agent Security
The Binance investigation adds another dimension to this discussion. According to the company's November 3 report, attackers used an internal AI tool as part of their effort to compromise additional engineers' devices. The report does not establish that an autonomous AI agent independently carried out the theft, but it demonstrates how AI-enabled tools can become part of an attack path when surrounding systems and access controls are compromised.
Whether assistants or autonomous agents, AI-enabled tools need clear identities, authorization boundaries, and monitoring appropriate to their access and capabilities.
As organizations bring agentic AI into development, infrastructure, security operations, and finance, the number of identities able to trigger privileged actions will keep growing. AI agents call APIs, retrieve secrets, modify infrastructure, execute code, and kick off business processes.
An attacker does not need to compromise the model itself. Compromising the identity, credentials, connected tools, or authorization context around an agent may be enough to turn its capabilities against you. AI agents should get narrowly scoped permissions, short-lived credentials, clear operating boundaries, extra authorization for high-impact actions, and continuous monitoring. The faster we automate privileged operations, the faster authorization and containment must run to keep up.
The Real Lesson: Assume Compromise, Contain Privilege
The Binance hack shows how far attackers can go when compromised identities reach trusted infrastructure and sensitive operations. Protecting credentials matters, but so does controlling what those identities are authorized to do.
Least privilege is not just about trimming admin rights. It is about shrinking the blast radius of a compromised identity through limited permissions, stronger authorization, and the ability to detect and stop dangerous actions quickly.
Attackers will get in. The only question that matters is how far your privileges let them go.
How Far Could an Attacker Get With One Compromised Identity?
A single compromised identity can enable privilege escalation, lateral movement, and unauthorized access to critical systems.
In Joseph Carson's Defender's Privilege Escalation Handbook, explore how attackers exploit privileged access and what security teams can do to limit their reach.
Ready to strengthen privileged access controls? Explore Segura™™ PAM
Incident updates
Updated November 4, 2026. This analysis incorporates Binance's initial incident report, the FBI's attribution, and independent reporting available as of this date. The investigation remains ongoing, and additional technical findings may change our understanding of the attack.

