Trust Identity Protocol | AI agents
Agent observability is the ability to see what an AI agent did, when it did it, on whose authority, and with what result. The Trust Identity Protocol (TIP) records each agent action as a signed entry on a shared network, so anyone you choose can check the trail and see whether a step was changed or removed.
A tamper-evident record of every action your AI agents take. Your data stays on your systems, and only its fingerprint goes on the network.
A tool call, a message, a file change or a payment request.
The agent signs a fingerprint of the input and output with its own key.
The entry joins the TIP network, linked to the step before it.
Auditors, regulators and customers check any step without access to your systems.
It is the ability to see what an AI agent did, when, on whose authority, and with what result. Tracing tools show engineers what happened inside a run so they can debug it. TIP adds proof, a record of every action that the organization running the agent cannot quietly edit afterwards, and that people outside the organization can check.
Logs live in a database the operator controls. Anyone with access can edit or delete a line, and nobody outside can tell. When an agent books the wrong flight, sends the wrong email or moves money it should not have, the people affected have to take the operator's word for what happened. A record on TIP is signed when the action happens and kept on a network of nodes, so a later change shows.
It is tamper-evident, which is the honest term. Nothing stops someone from changing their own copy of a log. TIP makes the change visible, because an edited entry no longer matches the signed fingerprint on the network, and a deleted entry breaks the link from the step after it.
Each action, whether a tool call, a message sent, a file changed or a payment requested, becomes a signed entry. It names the verified organization responsible for the agent, the agent and its version, the time, and a fingerprint of what went in and what came out. It also links to the step before it.
No. Prompts, tool inputs, outputs and customer data stay on your systems. Only a fingerprint goes on the network. A fingerprint is a short code computed from the data. It changes if the data changes by a single character, and nobody can rebuild the data from it.
The agent signs with a key issued to it by the organization that runs it, and that organization holds a verified TIP-ID. When a person approves a step, such as a refund above a set limit, the approval is signed separately with that person's own fingerprint or Face ID. Human decisions and agent decisions stay apart on the record.
For years. Signatures use ML-DSA-65, the post-quantum signature standard NIST published as FIPS 204 in 2024, so records stay checkable even against future quantum computers. Fingerprints use SHAKE-256.
On the TIP network, a ledger built as a directed acyclic graph (DAG) and kept by a network of nodes. Each node checks the signature on every new record before accepting it. Organizations with high volume can run their own node.
| Field | What it shows | On the network |
|---|---|---|
| Responsible organization | The verified TIP-ID of the organization that runs the agent | Yes |
| Agent | Its name and version, signed with the agent's own key | Yes |
| Time | When the action happened | Yes |
| Input and output | A fingerprint only. Prompts, tool data and customer data stay on your systems | Fingerprint only |
| Previous step | A link to the record before it, so a missing step shows | Yes |
| Human approval | A separate signature made with the person's own fingerprint or Face ID | When a person approves |
Compare your log with its records on TIP. Every entry carries a fingerprint and a signature. If an entry was edited, its fingerprint no longer matches. If an entry was deleted, the next step links to a step that is missing. Either way the change shows, without anyone having to trust your database.
Every record names the agent and its version and is signed with that agent's own key, which the responsible organization issued under its verified TIP-ID. When several agents work on one task, each signs its own steps, so you can follow the task from agent to agent.
You replay the trail. The records show which agent acted, what it received, what it returned, which person approved the step, if anyone did, and when. That turns an argument about what happened into a check anyone can run. TIP shows what happened. Stopping a bad action and judging whether the agent was right stay the job of your guardrails and your people.
No, and no record system can. TIP shows afterwards exactly what the agent received and what it did, which is how you find an injected instruction and prove where it came from. Use it alongside the controls that block attacks, such as input filtering, narrow tool permissions and human approval for risky steps.
Yes. You choose who sees which records. An auditor, a regulator, a customer or a court can check any step against the network without access to your systems and without having to trust your database.
It supports the record keeping the Act asks for. Article 12 requires high-risk AI systems to record events automatically over their lifetime, and Articles 19 and 26 require providers and deployers to keep those logs for at least six months. TIP gives each logged action a signed, time-stamped record that shows it was not changed. It is evidence for compliance, not a certificate of it.
No. Keep the tracing tools your engineers use to debug. They show what happened inside a run. TIP adds what they cannot, an independent record that shows the trail is complete and unchanged, which matters when someone outside your team has to rely on it.
It proves who acted, when, and whether the record has changed since. It does not prove that the agent's decision was right, that its output is true, or that the person who approved a step read it carefully.
Teams that let AI agents act for customers or staff, in finance, healthcare, insurance, legal services, travel, customer support and government, and anyone who will one day have to show what an agent did.
Write to [email protected]. Agent observability on TIP is a commercial service. We set up your organization's TIP-ID and agent keys, connect your agents, and agree volume and pricing with you. Teams with high volume can run their own node.
Agent observability on TIP is a commercial service for organizations that run AI agents. Tell us what your agents do and how many actions they take, and we will set it up with your team.
Read the standard, the Trust Identity Protocol. Want the agent itself certified? See the Global Seal of Trust.
Go deeper
