What happened
Resemble AI launched RAD-S1 (Agent Detection – System One) around Oct 30: a behavioral model that classifies each browser session as Person or AI agent by analyzing pointer movement, click timing, scrolling, touch and typing rhythm from a ~25KB script, with scoring on Resemble's servers. In an internal benchmark across 2,000+ agent visits and 12 stacks, it caught 98.8% of disguised agent visits versus <=14.8% for six incumbent bot-detection tools, and it returns a verdict (allow/redirect/coach) rather than auto-blocking.
Why it matters
It provides the first-order signal organizations need to govern agentic traffic — knowing whether a signed-in visitor is a person or an agent — addressing a gap highlighted by Schwab, JPMorgan and Amazon restricting outside AI agents; it can feed anti-abuse, shadow-agent and browser-agent governance workflows.
Applicability
Fraud/anti-abuse, web-security and agent-governance teams; lightweight one-script integration, but treat as one signal among controls and calibrate the ~5-6% human false-positive rate.