Insight for: Show HN: We post-trained a model that pen tests instead of refusing

An AI model and harness for penetration testing and security scanning, post-trained on CTF contests.

Analyzed: Jun 21, 2026

This product directly addresses a critical market gap: accessible, un-guard-railed AI-driven penetration testing for SMEs and mid-market. Current LLMs are either restricted or too generalized, leaving these segments vulnerable. By post-training on CTF data, the solution offers practical, exploit-driven vulnerability identification, moving beyond "vibes-based findings" to verifiable exploits. The CLI-based, local code scanning with context sent over TLS to an inference API balances security with utility. The "Pen test" mode, though gated, promises active adversarial testing in sandboxed environments, a significant differentiator. This targets a high-value problem: proactive security for underserved markets. The pricing model (free scan up to 2M tokens, then paid) lowers adoption barriers. The inherent safety concerns of such a powerful tool are acknowledged, indicating a strategic approach to responsible deployment.

post-trained model pen tests guard-railed offensive tasks cyber-focussed models SMEs mid market vulnerabilities AI cyber tools wrappers foundation model capture-the-flag contests Security scan read-only audit local codebase Pen test active adversarial mode live system sandboxed environment exploit request response integer overflow transfer path auth and secrets issues Bank of Anthos Kimi K2.6 (open weights) SFT on CTF writeups RL with verifiable rewards exploit checks multi-agent swarm orchestrator subagents CLI local binary inference API TLS tcpdump tokens Cosine (YC W23) domain verification

Hacker News Post

Parent Entity

Show HN: We post-trained a model that pen tests instead of refusing

Score: 77