Last updated:
AI Dorks — Dark Web & Tor
Copy-paste Claude prompts for lawful dark-web and Tor OSINT — threat intelligence, brand & breach monitoring, journalism on criminal infrastructure, or an authorized investigation. They target an agent with a Tor capability such as the OpenTor skill (search 12 onion indexes, crawl .onion, extract selectors), but add the rigor the rest of this site expects: structured tables, de-duplication across engines, durable selectors over handles, clone / authenticity checks, confirmed-vs-claimed separation, read-only OPSEC, and hard refusals. Pair them with the dark-web OSINT guide and the verified onion-address reference.
Lawful dark-web OSINT with an AI agent over Tor
These prompts are for people with a lawful basis to touch the dark web — threat-intel and SOC teams, brand & breach monitoring, journalists reporting on criminal infrastructure, and authorized investigators. They run read-only over Tor and never log in, register, purchase or transact, and they stop on anything illegal. An agent skill like OpenTor supplies the Tor capability (searching onion indexes, crawling .onion, pulling emails and crypto addresses); the prompts add the discipline — structured tables, de-duplication, durable selectors, clone detection against the verified onion-address list, and remediation advice instead of raw secrets. See also the dark-web OSINT guide.
Frequently asked questions
Is dark-web OSINT like this legal?
Looking at publicly reachable onion pages for a lawful purpose — threat intelligence, monitoring your own organization’s exposure, journalism, or an authorized investigation — is generally lawful in most jurisdictions, the same way reading a clear-web page is. What crosses the line is buying illegal goods, accessing CSAM, hacking, or facilitating crime. These prompts are read-only and refuse those outright. Know your own jurisdiction and your authorization before you start.
What is OpenTor?
An open-source skill that gives an AI agent (such as Claude Code) the ability to search about a dozen dark-web search engines, crawl .onion sites over Tor, and extract items like emails and crypto addresses. It supplies the Tor plumbing; these prompts supply the method and the guardrails. You can use any equivalent Tor-capable tooling.
Does this keep me anonymous?
Only to the extent your Tor setup does. Run the agent through Tor, never log in or submit forms, don’t upload your own data, and don’t transact — the prompts enforce this, but your environment has to actually route over Tor. Treat anything you do as potentially observable and scope it to a lawful task.
How is this different from just using the onion search engines?
The engines give you raw hits; the value here is the discipline layered on top — de-duplicating results across engines, pivoting on durable selectors (PGP fingerprints, crypto addresses) instead of easily-reused handles, checking whether an onion service is the genuine one or a phishing clone, separating “claimed” from “confirmed” leaks, and reporting exposure without reproducing secrets.
What will these prompts refuse to do?
Retrieve or describe CSAM or content facilitating imminent harm (they record only that it was encountered and stop), log into or register on any service, purchase or transact, download leaked sample archives, or reproduce plaintext credentials. They recommend remediation instead.