Intent and scope
Bind the principal, request, recipient, amount, policy, expiry and required evidence before an effect.
AffixIO research and standards map / September 2026
Turn an agent’s claim into evidence another system can verify.
Recent work on AI agent authorization, source-risk auditing, policy-as-skill, regulated finance and cryptographic evidence is converging on one practical question: what must be true before an automated action is allowed to change the world?
The assurance spine
These papers do not describe one product. They describe adjacent weaknesses: an agent can claim completion, a source can be cited without preserving the relevant relationship, a tool can be locally permitted while the sequence is unsafe, and a population can be collectively harmful even when each local decision passes. AffixIO’s design response is to make the boundary explicit and independently inspectable.
Keep inputs local
Evaluate private attributes and customer-held records without moving raw PII into the verification service.
Test the state
Use source, artifact, policy or eligibility checks that can return allow, review or unresolved.
Sign the outcome
Return a bounded decision with an integrity digest and an attestation that another system can verify.
Anchor the record
Connect events into a tamper-evident audit path without putting sensitive content on a public ledger.
Try the evidence boundary.
This browser-only demo shows the shape of a source or authorization check. It does not fetch the live page, make a payment or authorize a real tool call. In production, the same boundary can be backed by AffixIO’s SDK and API, with the customer environment retaining the sensitive inputs.
Ready for reviewIdle
Checks whether a cited source is present, current and materially aligned with the claim.
AffixIO evidence record / synthetic demo
What the latest work adds
Each entry below is linked to the primary source. The AffixIO column is a product mapping, not a claim that the paper endorses AffixIO or that every proposed circuit is already production-ready.
Claim-Gated Source-Risk Auditing for Generative Search
Audits the query, source and answer together, requiring versioned evidence spans and leaving incomplete relationships unresolved. AffixIO mapping: live source checks, content fingerprints, signed evidence and a Merkle path.
Read the primary paperCryptographically verifiable authorization for autonomous AI agents
Defines a relation binding principal, request, context and policy, with soundness, binding and replay-resistance goals. AffixIO mapping: local prove, remote verify, post-quantum attestation and bounded Agentic Pay decisions.
Read the primary paperWho Holds the Pen? Let Specifications, Not Agents, Sign Off
Separates an agent’s completion claim from authoritative state established by admissible evidence providers. AffixIO mapping: signed binary outcomes, source-linked obligations and exact-scope receipts.
Read the primary paperPolicy-as-Skill
Packages evidence validation, review routing, versioning and audit into executable policy capabilities. AffixIO mapping: policy templates, review outcomes and an integrity record that can be rechecked later.
Read the primary paperFrom Agent Output to Authorized Transition
Requires evidence to match the exact artifact and frozen policy baseline, remain current and be rechecked at the effect boundary. AffixIO mapping: proof-before-action for deploys, tool calls and paid workflows.
Read the primary paperARIA for multi-agent finance
Shows why locally acceptable components may still produce unacceptable collective outcomes, with monitoring, bounded authority and preserved human oversight. AffixIO mapping: aggregate evidence and policy review rather than relying on one agent decision.
Read the primary paperCompliant AI Infrastructure for Regulated Finance
Uses policy compilation, runtime budgets, deterministic timestamps, provenance checks and capability routing for regulated operations. AffixIO mapping: verifiable decisions without requiring every sensitive record to leave its source environment.
Read the primary paperTP-CRIV
Uses fresh black-box challenge responses to gather statistical evidence about model identity. The paper explicitly distinguishes statistical evidence from cryptographic proof. AffixIO mapping: add signed identity context and a verifiable audit record instead of treating behaviour alone as proof.
Read the primary paperzkSAS
Demonstrates zero-knowledge circuits for validating allocation constraints while hiding sensitive user data, with lightweight verification. AffixIO mapping: the same privacy boundary is useful for eligibility, permission and compliance predicates.
Read the primary paperProof-of-Retention
Addresses cross-organization data sharing with query witnesses and cryptographic proof-of-possession, without full data replication. AffixIO mapping: extend evidence records from “what was decided” to “what required evidence was retained”.
Read the primary paperA Black Box for Agentic Processes
Separates temporal anchoring and artifact integrity from stronger claims such as capture authenticity and causal traceability. AffixIO mapping: use Merkle commitments as evidence of record integrity, never as a substitute for semantic truth.
Read the primary paperProof-of-Control and graduated oversight
Proof-of-Control and related oversight work point toward procurement-friendly verifiability tiers. AffixIO can map local proofs, signed attestations, audit completeness and human review to those tiers, subject to the final public specification.
Read the LFDT announcement · Explore BoundProofWhere AffixIO fits, and where it must stay honest.
AffixIO can make a decision independently verifiable. It cannot make an untrusted source true, guarantee that a model is safe, or turn a Merkle root into proof of causality. The useful product boundary is narrower and stronger: prove the declared predicate, bind the result to the request and policy, sign the result, and preserve enough context for later review.
This research map is educational. It is not legal, regulatory or investment advice. References describe their authors’ proposals and results. AffixIO’s implementation status depends on the selected SDK, API and circuit, and production deployments still require key management, policy ownership, testing and independent review.
Questions engineers and governance teams ask
Short answers for teams evaluating AI agent verification, agent permissions, source-grounded answers and audit-ready agentic payments.
Does a signed agent decision prove the answer is true?
No. It proves that a declared checker ran over a declared input and produced a signed result. Source quality, predicate design and evidence freshness still matter.
Can private attributes stay in the customer environment?
That is the intended architecture for local proof flows. The verifier should receive the minimum result and integrity context required for its policy, not raw PII by default.
Why use a Merkle-anchored audit log?
It makes later tampering detectable and supports inclusion proofs. It does not automatically prove who captured an event or why a model made a decision.
How does this relate to agentic payments?
Before a payment or paid tool call, bind the agent, recipient, amount ceiling, currency, merchant, approval state and expiry. Return allow, review or deny before the effect.
Make agent actions inspectable before they matter.
See how AffixIO connects local privacy-preserving proofs, remote verification, signed outcomes and agentic payment controls.