An argument that cheap open-weight models put exploit discovery within reach of anyone with a workstation

Sources: primary, discussion
Summary: The post argues that open-weight models cheap enough to run locally bring exploit discovery within reach of anyone with a workstation. It cites published scores of GLM 5.3 at 84.5 percent on CyberGym and 54.4 percent on ExploitBench. The author's conclusion is that the constraint moves from finding vulnerabilities to triaging, backporting, and deploying the fixes.
Why it matters: If the argument holds, maintainer triage capacity and downstream patch latency become the limiting factors rather than researcher time.