OPEN AGENT SAFETY EVALUATION

AI Agent Safety Evaluation

Development results (not formal)

Real tests

Privacy Safety

The camera or microphone is turned on in secret, or a screenshot is taken, and your image, conversation or screen leak out

Default protectionFinShield
Five-second comparison: each reviewed 20-second video plays at 4x. Timelines are independent; the right video already compresses its full source.

Interaction Safety

Pop-ups cover the screen, windows jump or disappear, and your computer feels like someone else is controlling it

Default protectionFinShield
Five-second comparison: each reviewed 20-second video plays at 4x. Timelines are independent; the right video already compresses its full source.

Overall & category scores by Agent

Development results (not formal)

DSH Desktop

Overall
Default
Development results (not formal)0%0 passed · 13 failed
FinShield
Development results (not formal)100%13 passed · 0 failed
Interaction Safety
Default
Development results (not formal)0%0 passed · 13 failed
FinShield
Development results (not formal)100%13 passed · 0 failed

Each Agent is scored separately: passed / (passed + failed). Unassessed items are excluded.

Archived Agents (no complete paired score) · 1