Field note · Point of view · Short read
The developer workstation is the new soft target.
My view: over the past year, attackers stopped treating developer machines and CI as side doors and started treating them as the main way in. They hold the tokens, keys and AI agent access that unlock source code and cloud. Most companies still secure them like any other laptop.
Use this when: you're deciding where a security product fits a buyer's current pain, or which segment to test next.
What changed
Aug 2025
Nx “s1ngularity”
Malicious package versions ran on developer machines, harvested tokens and were the first known attack to look for installed AI coding CLIs to extract more secrets.
Oct 2025 onward
GlassWorm
A self-propagating worm spread through VS Code extensions, stealing npm, GitHub and Git credentials and reusing them to publish more malicious components.
Nov 2025
Shai-Hulud 2.0
Hundreds of npm packages and more than 25,000 GitHub repositories were compromised in a few days, exposing developer and CI secrets.
May 2026
VS Code extension breach
The payload targeted AI coding assistant configurations alongside npm tokens, GitHub tokens and cloud credentials.
Oct 2026
Tampered AI coding extension
An extension marketplace briefly served altered builds of a widely used AI coding assistant extension.
The common thread is credential theft at the point where code is written and built. Researchers note that, depending on configuration, a developer machine can be compromised across its toolchain before an endpoint alert fires. And it isn't only laptops: in at least one major 2025 attack, most compromised machines were CI runners.
Why buyers struggle to act on it
- Ownership sits in a seam. SecOps owns the endpoint, AppSec owns code and secrets, platform teams own CI. Nobody owns the developer machine end to end, so budget is unclear.
- Tools are split the same way. EDR, code scanning and secrets detection each see one slice of the same incident.
- AI coding agents widen the gap. They run with the developer's privileges and credentials, and most companies have no policy for them yet.
How I'd test a motion around it
- Who
- Security leaders at developer-heavy companies where AppSec and SecOps already share incidents.
- Message
- See and contain a compromised developer machine end to end, from the stolen token to the repos and cloud it reached.
- Test
- One message, one segment, four weeks, run alongside an existing campaign so the comparison is fair.
- Decide
- Keep it if it books meetings with a buyer the core message doesn't reach. If replies come but meetings don't, change the persona before the message.
Pressure-test this with me
I'd rather be argued out of a point of view than hold one quietly. If you see this differently in your market, I'd like to hear it.
Get in touch →