# RFC: Human delegation provenance for LangGraph multi-agent chains

**URL:** <https://forum.langchain.com/t/rfc-human-delegation-provenance-for-langgraph-multi-agent-chains/3509>\
**Category:** Talking Shop\
**Created:** [April 20, 2026, 1:37pm UTC](https://forum.langchain.com/t/rfc-human-delegation-provenance-for-langgraph-multi-agent-chains/3509 "2026-04-20T13:37:35Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Siri](https://avatars.discourse-cdn.com/v4/letter/s/a183cd/32.png) [@Siri](https://forum.langchain.com/u/Siri)\
**Post date:** [April 20, 2026, 1:37pm UTC](https://forum.langchain.com/t/rfc-human-delegation-provenance-for-langgraph-multi-agent-chains/3509/1 "2026-04-20T13:37:35Z")

</div>

```auto
Hi LangChain community,

Multi-agent chains in LangGraph raise a question that's hard to answer today: 
when a sub-agent takes an action, which human authorized it, and with what scope?

I've been working on HDP (Human Delegation Provenance Protocol) a 
cryptographic chain-of-custody spec for agentic AI. A human signs a 
token with Ed25519, agents extend the chain as they delegate, and any 
node can verify offline (no registry, no network call).

npm: @helixar_ai/hdp
Spec draft: https://datatracker.ietf.org/doc/draft-helixar-hdp-agentic-delegation/
Repo: https://github.com/Helixar-AI/HDP

Would LangGraph benefit from a first-party HDP integration? Happy to contribute 
a LangGraph callback handler that auto-extends the delegation chain on 
node transitions. Curious what the team thinks.

```

---

<div class="post-metadata">

**Author:** ![sakunaj](https://yyz1.discourse-cdn.com/flex007/user_avatar/forum.langchain.com/sakunaj/32/2872_2.png) [@sakunaj](https://forum.langchain.com/u/sakunaj)\
**Post date:** [April 21, 2026, 6:07am UTC](https://forum.langchain.com/t/rfc-human-delegation-provenance-for-langgraph-multi-agent-chains/3509/2 "2026-04-21T06:07:04Z")

</div>

Hey, @Siri nice and important work for ensuring Agentic Security. I created the following demo by integrating langgraph and HDP!

Check it out: [Agentic AI security with HDP - Ragatouille](https://sakunah.com/ragatouille-book/hdp/)