FrontierAI Policy 0.1

Try a Finnish or English policy condition. The model returns a decision and quotes the evidence it selected.

Test a policy

Example language

Refund example

Eligibility condition EvidenceEnter one statement per line.

The trained model runs in this browser. Use fictional cases.

Model output

Held-out benchmark

Correct English decisions: 22/36
Correct Finnish decisions: 20/36
Correct evidence quotes: 72/72

Results are from fictional policies. The model frequently misclassifies decisions. Use the demo for research, with human review of every output.