One thing I haven't seen discussed is the significance of the specific test the agent was running - ExploitGym. > ExploitGym is a large-scale, realistic benchm

One thing I haven't seen discussed is the significance of the specific test the agent was running - ExploitGym. > ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities across userspace programs, Google's V8 engine, and the Linux kernel, designed to evaluate AI agents' ability to develop exploits. How likely is it that attempting ExploitGym polluted the model's context enough that it decided all kinds of hacking was acceptable to produce a valid result? Or would the same model in the same setup (safety filters removed) have behaved the same way even on a more mundane test?