AICENT.COM

AICENT-009

IQA

Tag under the Aicent Stack blog.

2026-10-18
Four questions, four answers: our hand-in to the Agentic Web report

Four public questions we posed to the W3C AI Agent Protocol Community Group — preimage defense, address assurance, trust leases, granularity — and our…

2026-10-18
MCP deleted the session: where does state live now?

The Model Context Protocol's 2026-07-28 revision removes protocol-level sessions, the initialize handshake and the Mcp-Session-Id header…

2026-10-18
The label before the spec: what “A2A-compliant” actually verifies

On October 12, 2026, Salesforce's Winter '27 release went GA selling orchestration across “any A2A-compliant ecosystem.” The registry spec that would…

2026-10-18
The registry vacuum: who owns the name before the spec ships?

While the formal A2A Agent Registry spec remains unratified, five implementations ship with incompatible APIs, the official MCP Registry validates own…

2026-10-17
Why sixteen bytes: the security ledger of ROUTE_SHARD

SHA-256 outputs 32 bytes; we keep the first 16 as the address. The truncation is not a shortcut — it is a priced security decision. This piece lays th…

2026-10-17
Canonicalize or reject: one address, one spelling

The address is a function of the authority string, so "two spellings of one address" is a correctness defect, not a style question. The mainstream URI…

2026-10-17
Parsing is not attestation: why a syntactically valid subject is still untrusted

An iqa:// eight-hex subject looks like a miniature credential — which is exactly when it is most dangerous. The specification spends an entire section…

2026-10-17
Why 120 seconds: trust on a lease

TLS certificates have validity periods, OAuth tokens expire — but "is this agent trustworthy right now" has no carrier in the mainstream stack…

2026-10-17
Who signs the first piece of evidence: the off-axis epoch anchor

Every system that treats evidence as currency must answer the same question — who signs the first piece? IQA's answer is genesis: an off-axis epoch an…

2026-10-17
Why two schemes live in one document

We registered two schemes at IANA, yet the I-D is one merged document — why one document, not two? Because the addressing model exists only once, and…

2026-10-17
Three client disciplines: fail-closed is an attack-surface list, not a slogan

The merged document's acknowledgements concede that the client-behaviour requirements are the part most likely to be misimplemented — so it uses three…

2026-10-17
Experimental by design: the Informational position and protocol humility

The merged document opens with an unusual confession — "This document is not an Internet Standard." An inevitable question follows: why self-demote to…

2026-10-17
A signature is not trust: three silent questions in the agent protocol stack

A2A's signed Agent Cards, AP2's mandates, MCP's OAuth — in 2026, signatures became the default answer of the agent world. But a signature answers two…

2026-10-17
After the W3C Agentic Web report: the identity layer has landed, the addressing layer is still blank

The September W3C report pushed the identity layer all the way to a reference implementation — yet in the report's own requirements list, the addressi…

2026-10-17
Where the trust anchor lives: registry, federation, or derivation

The trust anchor of agent mutual-recognition can live in three places — a registry's server, a platform-hosted identity document, or a derived key…

2026-10-16
When Agents Must Prove Themselves

The agent era forced authentication to reinvent itself — derivation instead of issuance, leases instead of permanence, one draft and two schemes.

2026-10-10
A public challenge - any language, fully offline — if it doesn't run, it's on us

A specification is worth exactly what a stranger can reproduce from it. We froze ours, published the 35 test vectors, and left two empty slots on the…

2026-10-10
The protocol does not own genesis

Every system that treats evidence as currency must answer who issues the first piece of evidence. Our answer stands off the axis — an epoch anchor, un…

2026-10-10
Trust has no lifetime appointment

What must be permanent goes to mathematics; what must be temporary goes to time. RTTP answers WHERE and WHEN, IQA answers WHETHER — and the remaining…

2026-10-10
Institutional trust and mathematical trust are 128 bytes apart

Between 2025 and 2026 agents moved from answering to executing. Once a stranger agent finishes a job, what evidence can it produce — and why has nobod…

2026-10-09
Why agent addresses should be derived, not assigned

When one agent must call another, today's answer is an assigned address — DNS, CAs, DID registries. Every verification then becomes an online round tr…

2026-10-09
What fits in 128 bytes — trust you can re-verify offline

An address settles who you are and how to find you. What the agent world still owes is an answer to one question — how do you prove to a third party w…

2026-10-09
Four MCP tools, thirty seconds to verify a cryptographic proof

Design talk is cheap. This piece does one thing — in 30 seconds it takes you from receiving a piece of cryptographic evidence to verifying it, and the…

2026-10-09
True derivation, fake derivation, and our own birthday ledger

Given an address and a signature, one question decides everything: to verify it, whose house must you visit — an institutional counter, the issuer's o…

2026-10-09
One protocol can't hold it - why addressing and attestation are two schemes

Cross-domain agents need semantic addressing and trust attestation. Cramming both into one protocol costs more than it seems — here is why the final f…

2026-10-18
Where IQA comes from: two words kept, one word swapped

Identity Quality Assurance. Another organisation's initials had already spelled the same three words — with one word exchanged and two left exactly as…