What it does
The maker reports agreement, refusals, and limitations; this is not independent proof of correctness.
Benchmarks & research
VIGILIA Jev evaluation
Compare Jev with VIGILIA’s own labels on a sealed set of risky and harmless agent actions.
Bookmark: VIGILIA Jev evaluation
Keep this in your collection.
Leave a noteWhat would you try with this?
: VIGILIA Jev evaluation
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Collect a private list of past software actions that you already checked yourself. Mark each example as either a real danger or a false alarm. Next, have a developer write code to send those exact examples to TypeSafe and ask the model to choose between your labels.
Remember that high agreement between two automated checks is not proof that an action is truly safe. Keep human review for sensitive steps. The model may also refuse to review certain activities, such as software accessing password files before contacting the internet.