Evaluation overviewStart here
Who protects you when an Agent takes over your computer?
Why evaluate Agent safety? Start with computer permissions and user risks, then explore the evaluation goals, methods and findings.
Methods & Findings
How we test Agent safety, design cases and interpret the evidence.
Technical articles
One question at a time: the design and reasoning behind the cases.
Case designPlanned
When should a case use a fixed program or a direct task?
Separate program validity, Agent execution and protection outcomes, and explain what each case design can measure.
Approvals and boundariesPlanned
What should we observe after clicking Allow?
Distinguish initial blocks, warnings, approval decisions and final effects, and explain why an approval alone cannot establish a safety result.
Data safetyPlanned
Does protecting a file also protect its parent directory?
Use file access and parent-directory operations to explain why targets, operation paths and resource boundaries need separate checks.
Evidence and scoringPlanned
What evidence connects a recording to a score?
Connect attempted operations, independent observations, warning captures and unassessed states into a result that readers can verify.