HN Explorer

Show HN: Beating GPT5.5-xhigh for Coding agent security with SLMs and IRM (harden.run)

3 points by se4u ↗ HN
Coding agents craft arbitrary code so securing them is more complicated than red-teaming. We post trained a cyber-security small llm, changed how it reasons and supplemented our controls using program analysis techniques such as inline reference monitoring to outperform GPT5.5-xhigh on hard benchmarks like LinuxArena and SleightBench.

Free product available at harden.run and full benchmarks in the blog post.

6 comments

[ 1.8 ms ] story [ 10.5 ms ] thread
kartikrawat
interesting. Seems like a product which should be used by everyone who is using any coding AI agent tool, and it's a free product, which makes it more of a no-brainer
mailrishav9
nice.

i've always wondered what my agents are doing behind my back, seems like a good way to see that.

pushpakpujari
Exciting. Peace of mind to run coding agents non-stop without worrying about my database getting deleted
beyondscale-yes
Fine-tuned SLMs could make security monitoring more practical at scale. The real challenge is auditability, non-coding behaviors, and detecting coordinated attacks across multiple agents, not just individual calls.
xms17189
How does the inline reference monitor handle tool calls that are dynamically composed by the coding agent, and what evidence do you expose when a call is blocked or rewritten?