Research lab under ZTASecurity
Measuring what local AI can do in security.
SpinlockLabs studies local large language models for security tasks. We run models on our own hardware, with fixed probes and published numbers. No marketing claims.
How we work
A project page states the question, the setup, and what has been measured. Findings land in the blog once numbers are stable.
-
Local by design
Models run on our own GPUs. No prompts or data leave the lab. The inference engine is open source.
-
Fixed probes
Every task has a defined rubric. The same probe set runs across model versions and quantizations, so results are comparable.
-
Failures included
We report how models fail, not only what they get right. Hallucinated CVEs and wrong patches are data.
-
Hardware named
Results carry the machine they came from: two Radeon AI PRO R9700 GPUs today, with the engine and model versions listed.
Questions about the lab?
We answer on contact@ztasecurity.com. For ZTASecurity services, see the parent organization.