Skip to content
IAM Insights

Explainers · Glossary

LangGraph

A LangChain framework for building stateful, multi-agent orchestration graphs.

reference →

In our briefings

The claim each of the newest briefings on LangGraph would have you leave with.

  1. ZeroID's delegation narrows authority at every hop, a mechanism we did not find in Uber's write-up. A sub-agent registers its own identity and keypair and signs an actor token; the orchestrator presents that token to the token service and asks it to delegate, and the service returns a token whose scope is the intersection of what the orchestrator requested for the sub-agent and what it held itself, keeps the full chain of actors in the token, and refuses to go deeper than a policy's maximum delegation depth. Revocation is the mirror: revoke any token in a chain and everything downstream is invalid at once, checked through a networked session call rather than local key verification. It is one implementation, published as code, from the vendor that wrote it.

    Highflame's open-source ZeroID narrows an agent's token at every hop, a step we did not find in Uber's write-upBI-FIND-2026-0137

  2. The guide's answer to tool misuse is to give each step of the plan only the tools it needs. The Planner may need no tools at all, and the Executor gets only the tool or tools the current step needs, so that reaching the email tool during a calculation step is meant to be "architecturally impossible". In the authors' own LangGraph code, the planner model names the single tool for each step and the Executor runs only that tool, so the limit comes from the Planner's own output, and the paper does not say what stops an injected or mistaken Planner from naming a broader tool. The authors' own threat table says Plan-then-Execute does not scope tools by itself, so the scoping comes from the frameworks: coded by hand per node in LangGraph, declared per task in CrewAI, where Task.tools overrides Agent.tools (a rule they cite to a community forum post), and set through agent capabilities or custom logic in AutoGen. On top of that they propose RBAC roles as a ceiling, assigned by the Planner at plan time and checked against a predefined permission map, and say CrewAI does not implement this yet; nothing says what limits the role the Planner picks. The paper does not say how an agent authenticates to a tool or which user's access it acts under; credentials appear only as placeholder API keys in its code examples.

    A guide to securing agent plans has the planner choose each step's tools and leaves out agent identityBI-FIND-2025-0361

  3. BlockA2A's abstract says empirical evaluations show it neutralizing prompt-based, communication-based, behavioral and systemic attacks on multi-agent systems. The section behind that sentence rests on a table matching ten known attacks, from jailbreaks to agent-in-the-middle, to mechanisms in the authors' defense engine; it reports no attack run against the prototype and no detection or false-positive rate. Its timing results come from one local Hardhat test chain and IPFS node, averaged over 10 runs: 64.0 ms for the slowest framework step, a state transition, and 135.0 ms to flag a Byzantine agent. Network effects were left out on purpose, and the authors say real transaction latency and network conditions "could be additional bottlenecks".

    A Tsinghua blockchain design for AI agents backs its defense claims with a table and reports no attack runBI-FIND-2025-0129

Cited in 4 briefings

Every briefing that discusses this term, newest first.

  1. Highflame's open-source ZeroID narrows an agent's token at every hop, a step we did not find in Uber's write-up

    Article · · Highflame (GitHub) · Sep 25, 2026

  2. A guide to securing agent plans has the planner choose each step's tools and leaves out agent identity

    Paper · · arXiv · Ron F. Del Rosario et al. · Sep 10, 2025

  3. A Tsinghua blockchain design for AI agents backs its defense claims with a table and reports no attack run

    Paper · · arXiv · Zhenhua Zou et al. · Aug 2, 2025

  4. RAS-Eval cites weak MCP authentication, yet maps none of its 3,802 agent attacks to an authentication weakness

    Paper · · arXiv · Yuchuan Fu et al. · Jun 18, 2025

Related terms

What appears alongside LangGraph in the same briefings.

CrewAIAutoGenLlamaIndexLangSmithMI9LangChainAgentic JWTin-toto