Blog · Research & Positions

The argument for per-action proof.

Technical writing on Execution Authority Infrastructure: fraud economics, AI agent authorization, regulatory analysis, and the architecture of per-action evidence. Every claim sourced.

Searches titles, summaries, and the full text of every article.

23 of 23 posts

AI AgentsArchitecture

The First Agentic Intrusion Was Fully Authenticated

The first disclosed intrusion run end to end by an autonomous agent produced more than 17,000 logged actions on valid harvested credentials. Every one of them authenticated. What that does to a credentialed audit trail, why attribution came from the operator, and the artifact that partitions the stream.

July 2026 · 8 min readRead
RegulationAI Agents

EU AI Act Article 14 Moved to December 2027. Here Is What Did Not Move.

On 29 June 2026 the Council of the EU gave final approval to the Digital Omnibus. High-risk obligations under Annex III, including the Article 14 human-oversight requirements, now apply from 2 December 2027. The Article 50 transparency obligations still apply from 2 August 2026. The requirement did not change. Only the date did.

July 2026 · 2 min readRead
AI AgentsFraud & Payments

Agentic Commerce Disputes: Who Holds the Evidence?

Agentic payment rails authorize the charge. None of them proves the mandated task. One dispute, walked end to end, and the artifact that ends the argument.

July 2026 · 3 min readRead
Regulation

DORA Article 9(4)(d): Evidence or Process?

DORA asks for strong authentication and protected cryptographic keys. Supervisory review asks a harder question: can the control produce evidence? A requirement-level reading.

July 2026 · 2 min readRead
RegulationAI Agents

EPCS Non-Repudiation and the Clinical-Agent Boundary

DEA EPCS asks for two factors and a prescriber of record. A shared workstation session carries neither. Per-script signatures, and where the clinical agent's authority ends.

July 2026 · 2 min readRead
Architecture

What Underwriters Can Verify

Cyber-insurance questionnaires collect attestations. A deterministic per-action control is a different class of answer: one an assessor can verify independently. What the insured brings to renewal.

July 2026 · 2 min readRead
AI Agents

Human-in-the-Loop Is a Checkpoint. Out-of-Band Is a Trust Boundary.

HITL frameworks decide where the agent pauses. They do not supply an approval channel with an independent trust root. An in-band approval inherits the compromise of the channel it rides.

July 2026 · 3 min readRead
Architecture

What You See Is What You Sign, on Mobile: The Two Halves of the Promise

WYSIWYS makes a promise with two halves: signing what was sent, and seeing what was signed. One half closes cryptographically. The other has an owner, and honesty requires naming it.

July 2026 · 3 min readRead
Fraud & Payments

How Much Banking Fraud Can Execution Authority Infrastructure Actually Close? A UK Finance 2025 Breakdown

Of the £629.3M UK banks lost to fraud in H1 2025, roughly half to 60% is closed by device-bound cryptographic authorization: deterministically on the unauthorized category, and as a sharpening signal on a subset of APP fraud. The full breakdown, backed by UK Finance's own published figures.

April 2026 · 10 min readRead
Fraud & Payments

Thirteen Fraud Scenarios and What Execution Authority Infrastructure Actually Covers

Scenario by scenario: the eight APP categories UK Finance tracks, plus five unauthorized-fraud patterns and the emerging AI-agent category. For each: what happens, what exists today, and whether device-bound authorization neutralizes it, reduces it as a signal, or is not applicable.

April 2026 · 10 min readRead
Fraud & PaymentsArchitecture

Signal, Not Solution: How Execution Authority Infrastructure Integrates With Your Fraud Stack

Execution authority infrastructure does not replace Actimize, Featurespace, or Sardine. It contributes a single binary field (human_verified = true | false) that converts probabilistic fraud scoring into deterministic closure wherever identity uncertainty is the dominant source of error.

April 2026 · 6 min readRead
AI AgentsRegulation

EU AI Act Human Oversight: Cryptographic Proof vs Approval Dialogs

Article 14 requires effective human oversight of high-risk AI systems. The industry defaulted to approval dialogs. Prompt injection, agent compromise, and consent fatigue prove that interpretation was never enough.

April 2026 · 3 min readRead
Fraud & Payments

APP Scam Fraud: Why OTP and Risk Scoring Can't Stop It, and What Actually Can

OTP and risk scoring cannot stop authorized push payment fraud. That is a structural fact, not a product gap. The honest question is which slice of the fraud market per-action cryptographic signing deterministically closes, and which slice it cannot. UK Finance 2025 data, scoped.

April 2026 · 5 min readRead
AI Agents

Prompt Injection: Why Dialogs, Filters, and Policy Cannot Stop It

Prompt injection is not a bug. It is a property of how language models process input. Defenses that live on the same surface the attacker controls eventually lose. The only defense that survives is orthogonal to the model.

April 2026 · 3 min readRead
RegulationFraud & Payments

PSD2 Article 97: Strong Customer Authentication, Dynamic Linking, Non-Repudiation

Article 97 demands three things: two-factor authentication, a dynamic link to amount and payee, and non-repudiable evidence. Most production implementations satisfy one of the three.

April 2026 · 3 min readRead
ArchitectureFraud & Payments

Cryptographic Receipts for Financial Transactions

Six fields, one signature, no single party in possession of the whole truth. The anatomy of the artifact banks will need to produce, and the one regulators will increasingly ask to see.

April 2026 · 4 min readRead
Architecture

Deterministic Human Authorization, Explained

Fraud scoring produces a probability. Deterministic human authorization produces an artifact. For high-stakes actions, the difference matters more than the math.

April 2026 · 3 min readRead
Architecture

Why Session Logs Are Not Proof (And What Is)

A log is produced by the platform. In a disputed action, it is the platform's own account of what happened. Proof requires an artifact the platform alone cannot write.

April 2026 · 3 min readRead
ArchitectureRegulation

Execution Authority Infrastructure: A Primitive for Regulated Systems

Identity, fraud scoring, and policy engines each solve a real problem. None of them produces the artifact regulated systems actually require: a cryptographic proof that a specific human authorized a specific action at a specific moment.

April 2026 · 3 min readRead
Architecture

Why Sessions Failed: And Why PSEA Had to Exist

Sessions inherited trust from login and extended it indefinitely. PSEA corrects this by requiring proof of authority at the moment of action.

February 2026 · 2 min readRead
Identity

Offline Authentication vs Cloud Authentication (2025 Guide)

A technical comparison of offline-first authentication and cloud-based identity systems, including security, privacy, availability, and attack surface.

December 2025 · 4 min readRead
Identity

Why Login Authentication Is Not Enough for Zero Trust

Login authentication verifies access, not intent or human presence. Here's why Zero Trust requires action-level verification.

December 2025 · 2 min readRead
Identity

Passwordless Authentication Still Doesn't Verify Humans

Passwordless systems reduce friction, but most still verify credentials, not real human presence.

December 2025 · 2 min readRead

Prefer the short answers?

The Learn section answers the questions buyers actually search, and the FAQ covers the rest in two paragraphs each.

Browse LearnRead the FAQ