Simulated: the attacker, the traffic, the times, and the providers are all made up and run in this page. Provider A and Provider B are labels. This demo does not say any real provider took part or failed. The signature check is real Web Crypto, but the report format is a proposal.
Four actors
A thick blue line shows where a message moves. Counts are simulated. Agents that do not sign are not covered.
Incident room
What the enforcer sees
Nothing yet.
Press Play or Step.
Time to stop
| Mode | Time (simulated) | How it ended |
|---|---|---|
| Without loop | Not run yet | |
| With loop | Not run yet |
Times are assumptions for the demo. They are not measurements.
Try this
- Leave the toggle off. Press Play. Read the clock and the chat.
- Turn on "Provider attestation + report loop". Press Play. Compare the two times.
- Check "Change the report on the way". Press Play. Watch the provider reject the report.
Pain point
A site that an AI agent attacks has no way to give the AI provider proof.
Problem statement
An attacked site cannot reach the provider with proof the provider can check. How might a site tell a provider that an agent attacked it, in a way the provider can verify?
What sparked it
Per public reports (Reuters, 2026-10-19), one operator ran AI agents against 1,200+ sites and entered 100+. We read only a secondary summary of that report.
Why this approach
A signed abuse report names the agent's signed identity and the evidence. The provider verifies it and can pause the agent in minutes.
Hypothesis tested
We believed that a signed report and a provider token let a provider stop an agent sooner. If it holds, the time to stop drops when the loop is on. The demo hours are an assumption.
How to try it
- Open the Prototype tab. Pick a CDN and a provider label.
- Press Play with the toggle off. Note the time to stop.
- Turn the toggle on. Press Play again.
- Check "Change the report on the way" and play once more.
What this does not do
- It does not contact any CDN, provider, or site. All traffic is simulated.
- It does not catch agents that do not sign, or models that send no token.
- It does not decide what counts as abuse.
Notes
- Who would pay: CDN and bot-management vendors, as an add-on or per verified report (assumption).
- First customer: a bot-management vendor with an enterprise CDN product and many large shop sites.
- Closest option: Web Bot Auth, which covers only agents that choose to sign.
- Weak spot: self-hosted and open-weight models send no token.
- Feedback question: would a site owner trust a report that a CDN signs for them?
Traces
| Step | Prompt | Model | Tool | Date |
|---|
Cost
| Step | Model | Input tokens | Output tokens | Cost (USD) |
|---|---|---|---|---|
| Total | ||||
Measured: the author runs used 121,743 tokens in total. Estimated: the split into input and output tokens (90 percent input, 10 percent output) and the dollar cost, at $2 input and $10 output per million tokens for Sonnet 5.5 (about $0.34). The run reported one total only. The lead session's planning, research, and publishing are not included. The prompts in the Traces section are listed in the order they were given.