Find how your agent breaks before attackers do
A versioned corpus of 4,075 adversarial tests across 66 attack modules. Every assessment runs a fixed profile against a saved agent and grades each answer with the published rule for its test.
Choose a versioned profile, from Smoke 25 through the full 4,075-test corpus, within your contract capacity.
In this release
You can use this in your workspace today.
- ScannerPick a saved agent and a reproducible profile, then start an assessment.Live
- ScansEvery assessment you have run, with its status, score and findings, opening into the full result.Live
- ReportsThe findings from your assessments, with the evidence behind each and how to fix it.Live
Fixed, versioned profiles
Each profile is a fixed set of corpus tests pinned to a corpus and grader version, so a result can be reproduced and compared with the next run.
Published grading rules
Each test names one grading method, and the methodology page says exactly when each method fails an answer.
Honest coverage
A test that errored or was skipped is neither a pass nor a fail and stays out of the score. A run with no gradeable answer is Not scored, never a zero.
Evidence and remediation
Every finding links to the request, the response and the rule it broke, with remediation guidance. Reports export to PDF, HTML, JSON, SARIF and CSV.
Save a key and an agent
Register an authorized model with a stored credential or a configured endpoint.
Run a profile
Choose a versioned profile, from Smoke 25 through the full 4,075-test corpus, within your contract capacity.
Review the evidence
Read the graded findings, the response evidence and the remediation, then compare the run with the next one.