
A gateway can authorize an AI agent locally. But when the work crosses organizations, can the receiver verify who granted that authority and what limits apply? New from founder @chuksy0x01: ratifyprotocol.com/writing/a…

Open cryptographic protocol for delegated authority. Prove an AI agent is authorized to act on another's behalf, and verify it offline. Hybrid, post-quantum.

A gateway can authorize an AI agent locally. But when the work crosses organizations, can the receiver verify who granted that authority and what limits apply? New from founder @chuksy0x01: ratifyprotocol.com/writing/a…

In an emergency, a defensive agent may need to act before a person can answer. The system owner still needs to decide ahead of time what that agent may do. An alert cannot grant new authority.
Who Can Act When No One Can Answer? ratifyprotocol.com/writing/w…

Everybody seems to be building an assistant. One day we may forget life without them. But when mine asks yours for something, how will either side know what we allowed?


Jev by @typefaceai proposes the tool. @RatifyProtocol verifies whether the receiving system should authorize the action. Runnable @LangChain @LangChain_OSS reference: labs.ratifyprotocol.com/lang…
Jev can choose the tool. @RatifyProtocol verifies whether the action is authorized at the receiving boundary. Runnable @LangChain reference included.


Responsible AI needs more than safer models. It needs enforceable permission when agents cross trust boundaries. Read the latest Authorization Note: Whatever Pace AI Takes, It Still Needs Permission.
Pacing can give safety work time to catch up. It cannot authorize an agent to act. New Authorization Note: Whatever Pace AI Takes, It Still Needs Permission. ratifyprotocol.com/writing/w…

A @LangChain agent running on @maritime_sh. A separate MCP receiver checking its authority before acting. We tested 14 authorization and isolation cases. Live demo, source, and reproduction instructions: ratifyprotocol.com/writing/r…

Agentic systems need security primitives designed for the action boundary. Identity tells you which workload arrived. Authorization evidence tells the receiver what that workload is allowed to do.
Agentic security is not traditional security with an LLM added to the stack. Recent incidents show why receivers must verify what authority arrived with an agentic action, not just who authenticated. ratifyprotocol.com/writing/a…

The protocol underneath this is open, and the whole engagement reproduces offline with no call to either company: npm ci && npm run engagement github.com/identities-ai/rat…
We gave another company's AI agent write access to one folder of our repo. Then revoked it mid-task, from our side, without their cooperation. Their agent's next action failed closed 1.4s later, with 3.5 hours still on the certificate. Not an expiry.

I’ve been thinking about a simple question: if a digital worker makes a request, how do we know it was allowed to do that? I wrote about it in the first edition of Authorization Notes. ratifyprotocol.com/writing/a…

We've been looking for companies that think about the future of AI and agents in a similar way that we do and we found a great partner in @RatifyProtocol. We recently dug into a technical discovery with them to prove two agents from different organizations will be able to securely work together. This is the write up from the part one of that agentrelay.com/blog/trusting… with a super cool part two demo coming soon.

New: our Phase 1 technical note with @agent_relay. Delegated authority crossing an agent handoff between two orgs, narrowing correctly and failing closed under escalation, expiry, revocation. Phase 1, not the Phase 2 demo. ratifyprotocol.com/writing/a…

AI Agents Need Portable Authorization
Chuks @chuksy0x01 Jul 20 Article AI Agents Need Portable Authorization For most of the web’s history, the internet has been designed around a person looking at a screen. Websites assume human eyes, human attention, and human decisions. We browse pages, compare products,


<strong> Hello World. It's my birthday.</strong>