Comment by Meleagris

19 hours ago

I’m surprised people don’t get refusals running this, or are they running it with cyber-enabled models like daybreak?

In my experience, even just hinting at reverse engineering to models from Anthropic or OpenAI leaves them extremely sensitive to refusals. After all, the same techniques used here can be used to find exploits.

Or are people using it with local models?

Curious.

I handed Astra Ghidra with some printer driver exes/dlls loaded, and a Windows 2000 VM with the software installed and said "reverse engineer this printer driver, it's hooked up and powered on on /dev/ttyUSB0, tell me when you're ready to print a test page". No refusals.

Reverse engineering and decompilation is not currently blocked by the models, leading to the proliferation of LLM-assisted decomps/recomps. If you ask it to find bugs you're more likely to get a refusal, but simply pointing an LLM at a Ghidra MCP server and telling it to trace program flow, rename functions, or answer questions behaves like normal.

Yes, it just rejected me once with Claude Code Opus 5.5, but it works fine with Codex 6.1-Sol (in my experience, it was the other way around)

I think people said that Claude refuses to work on RE sometimes.

I've had no issues with OpenAI.

  • Ironically, I've had the opposite experience. My OpenAI account was permanently banned for doing too much RE work, meanwhile the worst I get on Claude is a downgrade to Opus 4.8.

Opus 5.5 hasn't given me much trouble with games, but for DRM and licensing checks I usually have to drop down to 4.8 or 4.6.

The Chinese labs aren’t constraining their models as much as Western labs do. That’s why HF used GLM. If you remember the early times with how Gemini (Bard then) and Claude refused almost everything, you’d despise the idea of guardrailing. (Its important in some areas though)