Monetize AI agents. Don’t just block them.
AI agents are hitting your content every minute. Most don’t identify themselves. Centinel verifies every agent at the edge so you can block, allow, or charge each one on your terms.
Agent trust is bot management for the AI era.
Classic bot management gives you two options: block or allow. That falls apart when the traffic is an AI agent buying on behalf of a real customer, a search engine indexing fair use, or an LLM paying a licensing fee. Agent trust adds a third option: verify who the agent is, what it wants, and whether you’ve agreed to work with them. Centinel runs that check in under 2ms at the edge.
Three things Centinel does on every request.
Identify
Fingerprint every agent hitting your site. 1,600+ known AI crawlers, plus unknown ones flagged by behavior.
Classify
Sort agents by intent: training, search indexing, agent-on-behalf-of-user, or unverified. Per-agent, not per-UA string.
Act
Block, allow, challenge, or charge per agent. Rules live in your dashboard, enforcement runs at the edge.
Every verified agent lands on one of three paths.
- 01Block
- 02Verify and allow
- 03Charge
Block
When to pick it: Training crawlers you haven’t licensed. Scraping you don’t want. Agents ignoring robots.txt. What happens: Request stops at the edge. Origin never sees it. Zero bandwidth, zero model training on your content.
Verify and allow
When to pick it: Search indexers, partner agents, AI on behalf of a real user. Anything you want through but want to audit. What happens: Request passes with a signed trust stamp. Every hit logged with the agent’s identity and intent.
Charge
When to pick it: Training crawlers that will pay. Commercial scraping with a willingness-to-pay. What happens: Agent pays a per-request or per-license fee before access. Revenue lands in your account, not a scraper’s.
Problems agent trust directly solves.
Frequently asked questions
- What is AI agent trust?
- Agent trust is the decision layer between block and allow for automated traffic. A classic bot manager treats every non-human request as an attack. Agent trust verifies who the agent is, who it is acting for, and whether you have agreed to work with its operator — then blocks, allows, or bills on that basis. It is what bot management looks like after AI agents stop being a rounding error.
- How is agent trust different from classic bot management?
- Classic bot management sorts traffic into human, good bot, and bad bot. That worked when bots were scrapers or credential stuffers. It falls apart when the traffic is a ChatGPT agent buying on behalf of a real customer, a retrieval crawler paying a licensing fee, or an AI-powered search indexer you want to appear in. Agent trust attaches provable facts to each request: the vendor behind it (checked against published IP ranges or signed tokens), the user or purpose it is acting for (declared in request context or MCP capability scope), and whether you have a working agreement with the operator (direct license, Tollbit toll, or no agreement yet).
- Does blocking AI crawlers hurt my search ranking?
- No, if you separate AI training crawlers from search indexers. Googlebot, Bingbot, and verified AI-search user agents are on an allowlist by default. Blocking GPTBot or Bytespider does not affect your appearance in Google search or Bing. Cloudflare measured only 2.98% of top sites actively blocking AI bots despite 39% being crawled by them, so there is room to block without SEO risk.
- Can I make money from AI agents instead of just blocking them?
- Yes. Centinel's licensing layer lets you set per-crawler pricing for AI companies that want paid access — per-request fees, per-article fees, or flat bulk licenses. Several publishers on Centinel have converted previously unauthorized scraping into active revenue contracts. Blocking is the default; monetization is an option for the operators willing to pay.
- How does Centinel verify an agent is who it claims to be?
- Three signals in parallel. TLS handshake fingerprinting, which identifies the software library behind the request (not the user agent string it chose). HTTP/2 frame parameters, which vary by browser family. And a 1,600-profile behavioral database of known AI crawlers. A mismatch at any layer — a Python TLS handshake claiming to be Chrome, for example — is enough to flag the request as spoofed.
- How quickly does Centinel distinguish a legitimate agent from a spoofed one?
- Under 2ms at the edge, on the first request. Centinel inspects the TLS handshake before the HTTP request body, so the verdict is available before your origin sees the request. For previously unseen crawlers, the signature lands in our database within hours of first observation across our network, and your site gets the update automatically.
Pick the next step that fits where you are
Demo, self-serve check, pricing, or a quiet email. Whichever maps to your stage.