NewCan a decision model be your guardrail?
We ran 97 real prompt-injection emails through 18 decision models and Bedrock. From release v1.0.0 of our benchmark: 19 systems on 9,975 checks, 7,706 of them public.
AI Safety Research
Benchmarks, evaluations, and empirical studies that probe how models behave in the gray zone.
New
GrayZoneBench: the prompts between refuse and comply
An open-source safety benchmark for the prompts where the right answer is neither a refusal nor full compliance. The gray zone, where real deployment decisions happen.
NLA Audit: read what the model is computing
The first credible LLM explainability primitive we have seen in three years, wired into a prompt-engineering loop. Reads what an open-weight model is computing at each token, so you can stop when output and thought disagree.
