A backdoor in a LinkedIn job offer
29 points by hoistbypetard
29 points by hoistbypetard
I asked the agent to review the codebase and flag anything suspicious
This sounds profoundly misguided, but I may just be misinformed. What is to stop these malicious repositories from influencing an agent? Even if it's "read only"?
I'm not so sure about just letting an agent in a read-only environment churn through tokens to do an initial security audit is the best knee-jerk reaction. This author using a specific type of harness that "finds anything suspicious", but elides valuable content on the specific approaches that were used besides this general concept:
Another takeaway is that reviewing the code with a read-only agent turned out more productive than reading it myself
The exact nature of this is bound to change, and it's good to call out this one instance. It reads similarly to "read the code before you run it", but instead this article reads as "trust an agent to not lie or omit anything before you run it", which probably needs some improvement for it to be thorough.
When they said the job market was difficult, I didn't imagine it would become hostile.