The record of trust for agentic work.
TELOS makes what your AI agents actually do visible and reviewable. On the paths you instrument, consequential actions are recorded as they happen and kept as tamper-evident evidence, checkable offline by anyone.
TELOS is forming the record of trust for agentic work, at runtime.
The record of trust is a durable artifact: a tamper-evident, hash-linked chain of signed receipts documenting what an agent did, scored against the purpose a human declared — checkable offline by anyone, without trusting the agent, the vendor, or us.
Prototype boundary: the standalone receipt is unsigned today. The published benchmark record uses a demonstration key. Production receipt signing is pending.
How do you know what an agent actually did?
Identity tells you who acted. Logs tell you what was called. Neither tells you whether the work stayed inside the job a human set. The record answers that, action by action.
Six questions, answered for every action.
PURPOSE
What was this for?
What it means: The human-declared purpose the agent was given. On instrumented paths, actions are scored against it, so you can see how closely the work matched the job.
SCOPE
Did it stay inside?
What it means: The declared scope of the work. The record shows whether the work stayed within it.
BOUNDARY
Did it reach a line?
What it means: The lines a human said to watch. When an action reaches one, the record notes it and the decision goes to a person.
TOOL
What did it touch?
What it means: The tools and systems the agent used. The record captures which ones the agent touched.
CHAIN
Can you prove the order?
What it means: Each receipt is hash-linked to the one before it. You can prove the order and show nothing was inserted or removed.
CODEBASE
Can anyone check the code?
What it means: The receipt format and verifier are open. Anyone can read the code and recompute the hashes offline.
Three teams want three things at once.
Agents raise the stakes on a tension every organization already feels.
The board wants speed
Put agents to work and move faster than the competition.
Security wants safety
Not get burned when an agent does something no one meant to allow.
Compliance wants proof
Be able to show, later, what happened and why.
The record resolves it: move fast, and keep proof.
The stream behind the glass is real.
The rows behind these windows are real: TELOS watching its own agent. Thirty hash-linked steps, and you can check every one in your browser.
Mapped to eleven governance frameworks.
TELOS maintains alignment mappings from its observation and attestation work to eleven regulatory and standards frameworks, with article and control anchors.
| # | Framework | Year | Scope | Anchors |
|---|---|---|---|---|
| 1 | IEEE 7000 | 2021 | Ethical system architecture | IEEE 7000-2021 |
| 2 | IEEE 7001 | 2021 | Algorithmic transparency | IEEE 7001-2021 |
| 3 | IEEE P7002 | 2022 | Data privacy engineering | IEEE P7002-2022 |
| 4 | IEEE P7003 | 2024 | Algorithmic bias | IEEE P7003-2024 |
| 5 | SAAI Framework | 2026 | Agentic AI safety (Watson & Hessami) | SAAI categories |
| 6 | EU AI Act (2024/1689) | 2024 | AI risk classification and obligations | Art. 9, 11, 12, 14, 15, 72, 73 |
| 7 | NIST AI 600-1 | 2024 | Generative-AI risk management | 12 GAI risks; GV / MAP / MS / MG |
| 8 | NIST AI RMF 1.0 | 2023 | AI risk lifecycle | Govern / Map / Measure / Manage |
| 9 | UC Berkeley CLTC | 2026 | Agentic AI risk profile (NIST RMF) | Govern, Map, Manage subcategories |
| 10 | NAIC Model Bulletin | 2023 | Insurance AI governance | Bulletin §3; Model #880 |
| 11 | OWASP Agentic Top 10 | 2026 | Agent security risks | ASI01-ASI10 |
These are self-assessed alignment mappings maintained in the TELOS governance review repository, not independent certifications. Mappings are being published progressively. TELOS governs alignment to declared purpose, not correctness of outputs, and does not certify legal compliance.
A complement, not a competitor.
ATF verifies who the agent is and what it may do; the record of trust attests what it actually did.
| Framework | Its lane | Relation to the record of trust |
|---|---|---|
| ATF | Zero-Trust front door: who the agent is, what it may do | Attests identity and permission. The record attests the action that followed. |
| AARM | Runtime enforcement layer | Acts on actions at runtime. The record observes and attests; it does not take control. |
| C2PA | Provenance for media content | Attests where content came from. The record attests what an agent did. |
| NIST AI RMF | Risk-management lifecycle | Asks for monitoring and accountability. The record supplies that evidence. |
| EU AI Act | Risk classification and obligations | Asks for oversight and documentation. The record is evidence toward both. |
Instrument once. The audit trail writes itself.
You add the record for one reason: to see what your agents do. The forensic history and the compliance evidence come out of the same signed receipts, at no extra step.
Three ways to check it for yourself.
telos-receipt · PyPI
Mint a receipt and verify it offline. No account, no keys.
pip install telos-receipt
v0.1.0. Runs on synthetic data, observation only. It is the receipt contract, not the calibrated engine.
GTC charter
The governance trust charter TELOS builds under. TELOS was selected into the cohort in 2026.
Read who reviews the work, and how, in the charter.
Review-repo bundle
The public review repository holds the framework mappings and verification material. Read it, and check it.
github.com/TELOS-Labs-AI/telos-governance-review
On our terms, plainly.
How is this different from ATF?
ATF verifies identity and permission: who the agent is and what it may do. The record attests what actually happened afterward. They fit together.
How is this different from AARM?
AARM is a runtime enforcement layer. TELOS observes and records. It does not take control of the work, and people stay in charge.
Is this like C2PA for agents?
Same shape. C2PA attests where a piece of content came from. The record attests what an agent did.
Does TELOS control what the agent does?
No. It observes and records. People stay in charge. When an action reaches a line a human set, the record notes it and a person decides.
Who is reviewing TELOS?
GTC evaluators and observers review the work under the charter. The review repository is public, so you can read the mappings and verification material yourself.
Start with one workflow.
Write down its purpose. Run your agent. Read the record.
Open to read, open to check.
Licensing
The receipt code is Apache-2.0. The review content and framework mappings are CC BY 4.0. Share and adapt them, with credit.
Selection and review
TELOS was selected into the GTC cohort in 2026 and builds under its charter. The review repository is public.
Published datasets
Validation datasets are archived on Zenodo with forensic audit trails, including the adversarial set (DOI 10.5281/zenodo.21629088) and the XSTest over-refusal set (DOI 10.5281/zenodo.21630399). The full table is in the whitepaper.
Author transparency
TELOS AI Labs Inc. builds and maintains this work. The framework mappings are self-assessed, not independent certifications.
What this does and doesn't claim
TELOS is observation first. The current TELOS internal deployment is passive: a read records what happened and does not block the underlying action. Coverage is scoped to the paths you instrument, so the record is not proof of coverage across every agent framework. Receipts are integrity-hashed and, in this prototype, unsigned; we do not claim that every operational event is TKey-signed. The record is evidence, not a certification or endorsement, and this page is honest about what current evidence does not support.