Building and studying AI systems for offensive and defensive security.
GLM 5.3 Flash post-trained with GRPO reinforcement learning on verified exploit trajectories from real vulnerabilities, released alongside an abliterated variant with refusal behavior removed for unrestricted use in security workflows.
Cantina
ReadGLM 5.3 Flash post-trained with GRPO reinforcement learning on verified exploit trajectories from real vulnerabilities, released alongside an abliterated variant with refusal behavior removed for unrestricted use in security workflows.
In a frozen three-month production cohort, Fix Review surfaced 97 issues across 62 of 193 fix pull requests. Here is what those patches missed and how Apex verifies them now.
With the right fix instructions, the strongest model fixed 90.4% of vulnerabilities correctly.
A same-codebase comparison of Apex, Claude Code Security, and Codex Security across validated findings, token efficiency, false positives, and operator overhead.
An empirical study of 4,032 machine-generated findings across 612 production web applications and APIs.
Cantina's AI bug hunter Apex analyzed 1,610 production runs across codebases, discovering predictable sub-linear scaling where doubling compute yields 40% more findings.
