A GitHub Issue Title Compromised 4,000 Developer Machines

A prompt injection in a GitHub issue triggered a chain reaction that ended with 4,000 developers getting OpenClaw installed without consent. The attack composes well-understood vulnerabilities into something new: one AI tool bootstrapping another.

@Migueldeicaza This isn't even funny. I mean it is, but it isn't at all, ya know? What have we done?
Cursor is rolling out a new kind of agentic coding tool | TechCrunch

Called Automations, the new system gives users a way to automatically launch agents within their coding environment, triggered by a new addition to the codebase, a Slack message, or a simple timer.

TechCrunch
@Migueldeicaza something something secure by design
@Migueldeicaza can someone pls explain for non techies like me?
@hagbard AI systems are being rolled out without much care, with catastrophic consequences.
@Migueldeicaza I'm gonna be saying "AI means read == execute" every day until an early grave, aren't I?

@slightlyoff @Migueldeicaza

That's a nice and easy way to sum up my issues with AI

@retrimental decades of chip, compiler and operating systems designers working more or less carefully on separating code and data in Von Neumann architecture computers.

Enter "AI": ahh, hell!

@slightlyoff @Migueldeicaza

@retrimental worst of it is dropping much more precise computing in favor of 8 bit floating point operations over huge matrices for inference, then calling it artificial intelligence. Yeah, doing flashy Budenzauber gets a lot of people to buy your knick knack.

@slightlyoff @Migueldeicaza

@Migueldeicaza
"Why existing controls did not catch it

npm audit: The postinstall script installs a legitimate, non-malicious package (OpenClaw). There is no malware to detect."

My POV is that OpenClaw is one of the most malicious programs ever distributed. Baffling that people deliberately install it!

Migrate from GitHub | GitLab Docs

Migrate from GitHub to GitLab.

@Migueldeicaza @slightlyoff Decades of “never trust user input” thrown out the window for AI…
@Migueldeicaza "prompt injection" is something I was not ready to process...
@Migueldeicaza almost like allowing anyone with a pulse to be in charge of software or touch a computer was not a good idea?
@Migueldeicaza
Why are we doing this to ourselves?!?!
@Migueldeicaza is this why none of my actions were deploying today?!
@Migueldeicaza What makes you think so? Only if you run a browser agent with system-level access? You gotta be pretty dumb & AI-pilled to do that. 🤔
@Migueldeicaza Tip of the icerberg. Not even the tip. A flaking off of one edge.
@Migueldeicaza Atm I can see some requests on /mcp and /sse HTTP paths here in my logs. Does this have something to do with this issue?