The Developer Tool Blind Spots Behind the Binance Heist

The $6 billion stolen from crypto shop Binance last month might be the biggest heist in history. And it all started from a compromised laptop.

Binance released its initial findings on Tuesday, describing a campaign that highlights a growing enterprise security challenge: attackers are increasingly targeting developer tools and internal AI agents to gain a foothold, move across environments, and access sensitive systems.

While the scale of the theft is extraordinary, the underlying security challenges are not unique to cryptocurrency companies. Three aspects of this incident are particularly instructive.

Software installation remains a critical point of exposure

The attack reportedly began with a familiar activity: an engineer on Binance's wallet team wanted to try a popular AI coding tool, searched for it online, and installed it using a tampered version of Homebrew.

This is a workflow that plays out across enterprises every day. Employees discover new tools, download software, and install applications or packages to get their work done.

The rapid adoption of AI has only accelerated this pattern. New coding assistants, extensions, and developer utilities are released constantly, and employees are often eager to experiment with them.

The challenge is that traditional software management approaches don't always extend to every installation method. Applications, package managers, command-line utilities, browser extensions, and IDE plugins can enter an environment through different channels, with varying levels of oversight.

As a result, security teams may lack a complete picture of what's running across their environments, let alone the ability to prevent risky or malicious software from being installed.

The lesson is that software control needs to extend across every installation channel.

Security teams need the ability to:

• Discover software regardless of how it was installed.
• Identify malicious, compromised, or unapproved software before it can introduce risk.
• Establish trusted software sources and enforce approval policies consistently.

Visibility is the starting point, but prevention requires the ability to act on that intelligence before risky software enters the environment.

Developer endpoints are high-value targets

A second part of the reported attack involved another engineer, referred to by Binance as "Engineer B," who received a coding assignment as part of a fraudulent job interview.

The assignment contained malicious code that was executed on a corporate laptop. According to reporting from The Register, Binance's endpoint security tools did not immediately detect the malware.

Developers have become particularly attractive targets for attackers. Their work requires them to download code, test new tools, and connect to systems that other employees may never access. Their endpoints often hold credentials, development environments, and connections to critical infrastructure.

Attackers know this and are increasingly finding ways to exploit everyday developer workflows like compromised software packages as we recently saw with the malicious Claude Code extension in Cursor's marketplace.

The lesson is that developer endpoints deserve greater scrutiny, not because developers are less careful, but because of the access they have and the software they routinely work with.

AI agents need guardrails of their own

Once inside, the attackers turned to Binance's own internal AI tool and used it to reach other engineers' computers, including the one holding the deployment token that kicked off the theft.

Binance hasn't said exactly how the tool made that possible. Any AI agent that can act on engineers' machines or call internal systems carries their access with it, and whoever takes over a user's laptop gets that reach as well.

Most organizations are adding these tools faster than they're adding guardrails:

• Coding agents can install their own extensions and MCP servers.
• People outside engineering now write and run code with AI's help.

Before an incident puts them to the test, every security team should be able to answer these questions about its AI tools:

• What machines, systems, and credentials can each tool reach?
• Which actions require a human to approve them first?
• Which agents and MCP servers are running on which endpoints today?

The lesson is that securing AI requires control over more than the tools themselves. Organizations also need to govern what those tools can access and what they're allowed to do.

How to reduce risk with AI-powered prevention 

The Binance breach shows how quickly a compromised developer environment can become a much larger problem. And with developers and AI agents adding software and connecting to new systems faster than ever, security teams can't rely on manual reviews to keep up.

The best defense starts with prevention: reducing the attack surface, stopping risky software before it has a chance to cause harm, and stopping internal AI agents from being a springboard to other important systems.

That means putting AI to work on the security team's side, too.

AI agents can continuously help security teams:

• Research software: Investigate every application, package, and extension for its source, known vulnerabilities, maintenance history, and publisher reputation.
• Manage policies: Help security teams decide what software should be approved, restricted, or blocked, and keep those policies current as new risks emerge.
• Enforce guardrails: Block malicious or unapproved software at supported control points before it reaches employee devices, and control what AI agents can access and do.
• Reduce the attack surface: Identify existing unvetted software and help security teams remove it before it creates further exposure.

This is where Glow comes in, putting AI agents to work to help security teams stay ahead of software risk instead of constantly catching up. By controlling what software is being adopted, what approved agents can access, and what actions agents can take inside the environment, Glow can turn developer tool blind spots into security control points.    

To learn how Glow can help you take control of your environment, talk to our team.