Resources
Learn how to make consequential AI actions verifiable.
For buyers, security teams and engineers. Curated, not a feed.
Start here · The problem
Your logs can be right — and still fail you when they're challenged.
The provider records the action, stores the record, controls the system behind it and later presents the evidence. Why that is the problem, and what independently verifiable evidence changes. Ten minutes.
Use case“I never authorized this.” Repudiation claims against AI agentsThe most common dispute, what is actually alleged, and what settles it.
GuideHow to add verifiable evidence to an AI-agent productFrom the first Environment to the first verification link in your product.
ArchitectureWhat Tallwright proves, and what it does notIntegrity and consistency, not truth or completeness. For a security reviewer.
Browse by purpose.
Three shelves. Each shows its three most useful pieces.
Understand
Why provider-owned logs are not independently verifiable evidenceCorrect is not the same as checkable.
Evidence before disputes, incidents or auditsTiming is the point.
Use cases: refunds, approvals, account changes, security actions, support and voiceWhere it becomes relevant in real products.
Why provider-owned logs are not evidence, and where that bites.
Implement
How to design evidence before a dispute happensWhich actions, what belongs in a record, what never should.
How to prepare for customer or security reviewThe questions you will be asked and the documents that answer them.
How to choose Cloud Proof vs Own Identity / Self-HostedA decision by role: keys, logs, operations.
Guides for teams evaluating or building verifiable evidence.
Architecture and trust
Data boundariesCommitments cross; payloads and subject identifiers do not.
The Witness, Environments and identitiesWhat a witnessed checkpoint is, and why evidence never mixes.
Independent verification and deployment boundariesPublic page, endpoint, offline CLI; Cloud Proof to Self-Hosted.
For CTOs, security teams and technical buyers.
Research and field notes
Original Tallwright thinking on agent accountability and evidence.
Informed category work, not commentary. Published as it is ready; the first three pieces are in preparation.
Trust and control asymmetry in AI-agent productsIn preparation
Why the party that benefits from a record is the party least able to vouch for it, and what follows for evidence design.Emerging evidence requirements for consequential agent actionsIn preparation
What auditors, insurers and regulators are beginning to ask for, and where current logging falls short.Verification patterns: who checks, when, against whatIn preparation
Customer-run, auditor-run and offline verification, compared by what each establishes.Ready to make it concrete?
Cloud Proof explains the product. Developers shows the integration.
Product and company updates are kept separately: Changelog · Status