---
title: "Whether US Incident-Reporting Rules Would Reach Operators in Lagos and London Depends on the Contracts Before the Statute"
canonical: "https://theupsidejournal.com/articles/us-ai-incident-rules-lagos-london-contracts/"
published: "2026-10-28"
column: "The Beltway Byte"
author: "Olusegun (Olu) Olufunmiloye"
subtitle: "The Artificial Intelligence Risk Management and Security Act of 2026 binds the parties it defines, and the text read for this article defines them by what they build or operate, not by where they sit."
description: "The US AI incident bill states no territorial scope. For Lagos and London operators on US frontier models, the contract is the likelier route to the duty."
---

# Whether US Incident-Reporting Rules Would Reach Operators in Lagos and London Depends on the Contracts Before the Statute

*The Artificial Intelligence Risk Management and Security Act of 2026 binds the parties it defines, and the text read for this article defines them by what they build or operate, not by where they sit.*

## Executive Summary

The Artificial Intelligence Risk Management and Security Act of 2026 binds the parties it defines, and the text read for this article defines them by what they build or operate, not by where they sit. It names "any developer or provider of a frontier artificial intelligence model" and operators of critical infrastructure that use AI in industrial control systems or other operational technologies. It contains no territorial clause, no definition of a "United States person", and no mention of a "deployer" or a "customer".

For a Lagos or London operator that builds on a US frontier model, the nearest route to the duty is therefore its vendor's contract; two further routes depend on definitions the bill leaves open. The bill has not passed.

## Introduction: A Question the Text Answers Only in Part

Coverage of the bill treats the incident-reporting duty as a US compliance matter, and the opposing view treats any foreign operator as simply outside it. Both assume a territorial answer that the text does not give. The [pillar analysis of the bill](https://theupsidejournal.com/articles/warner-ai-safety-board-frontier-model-access/) sets out what it requires and who sits on the proposed Board; this article starts from a narrower question: which party does each duty bind, and by what route would an operator outside the United States meet it?

Every statement about the bill below comes from its text as published by Sen. Mark Warner, one of its sponsors. Where the text is silent, the article says so, and an inference is labelled as The Upside Journal's reading. The article names no company in any hypothetical and offers no legal conclusion; operators with exposure should take advice from counsel qualified in the relevant jurisdiction.

## Which Party Does Each Duty Bind, According to the Bill Text?

Section 5(a)(3) of the bill is the incident duty. It requires reporting of "any confirmed artificial intelligence safety incident or artificial intelligence security incident" within 30 days of confirmation, and within 72 hours where the incident poses an imminent threat to national security, critical infrastructure or public safety. Two categories of party carry the obligation. The first is any developer or provider of a frontier artificial intelligence model. The second is any operator of critical infrastructure that utilizes an artificial intelligence model, in the context of managing industrial control systems or other operational technologies.

The definitions decide how far those categories extend. Section 2 defines a **frontier artificial intelligence model** as "an artificial intelligence model, or system combining multiple artificial intelligence models" that "exhibits or could be modified to exhibit high levels of performance at tasks that pose a serious risk". A **developer** is "a developer of a frontier artificial intelligence model". "Provider" is used in Sections 3 and 5 but is not defined in Section 2, and neither is "confirmed". Section 5(a)(3)(B) applies the duty whether or not the model is, or is intended to be, publicly available, and whether the incident arises in development, training, testing, evaluation, red-teaming, deployment or operation.

The financial exposure sits with the developer; the [incident-clock piece](https://theupsidejournal.com/articles/ai-incident-clock-vendor-contracts/) carries the penalty figure and its Section 3(c) caveat.

The verdict is narrow: the duty binds those who make or supply a frontier model and those who run covered infrastructure with one, and the enterprise that merely builds a product on top is not named.

## Does the Text State Any Territorial Scope?

It does not, as read. The bill has no territorial clause and no definition of "United States person". The words "deployer" and "customer" do not appear. The bill names the United States in Section 5(a)(4)(B) ("impact to the people or economy of the United States") and Section 5(a)(4)(C) ("widely used in commercial or public sector contexts in the United States"); neither states whether the duty reaches a foreign operator. Beyond those, the only foreign references are in two places: Section 3(a)(2)(C), under a heading on international participation, which refers to "foreign partners", and Section 3(f), which requires a developer to give the Board access to a frontier model not later than 45 calendar days before it introduces the model into "interstate or foreign commerce".

The contrast with other regimes is instructive. The [analysis of the EU AI Act](https://theupsidejournal.com/articles/eu-ai-act-2026-us-tech-african-fintechs/) shows a statute that says in terms that it applies to organisations whose AI systems serve European users, which is why a Lagos or Nairobi company can fall in scope once European customers are involved. The US bill makes no comparable statement. Silence cuts both ways: it does not extend the duty to a foreign operator, and it does not exclude one whose product meets a definition.

The verdict: reach, if it exists, has to be built from the bill's definitions, because the bill never states it.

## By What Route Would a Lagos or London Operator Meet the Duty?

Three routes are available on the text. The first follows from the contract. The second and third depend on definitions and are labelled here as readings.

### Route one: does the duty arrive as a contract term?

If the duty binds the US developer, a Lagos or London operator that licenses the model meets it as terms in the licence, not as an obligation of its own. This is the contract flow-down mechanism analysed in the [companion piece on the incident clock and vendor contracts](https://theupsidejournal.com/articles/ai-incident-clock-vendor-contracts/), which argues that the clock would move risk into four negotiated terms: notification, the definition of confirmation, audit and information rights, and indemnity. That reading is a hypothesis about how markets would respond; the bill does not mandate any contract term.

Published terms give a baseline. OpenAI's Services Agreement, effective 1 January 2026 and last updated 1 December 2025, was read for this article. The page contains no clause requiring the vendor to notify a customer of a security incident. It provides that, on written request and no more than once a year, the vendor supplies its most recent independent audit reports, which are treated as confidential information, and it gives the customer no right to run its own audit. The agreement incorporates a data processing addendum that the page does not reproduce and that may hold further terms. One vendor's terms are not a market survey; they show how notification is stated in one agreement today.

The economic consequence follows from that gap. If a notification obligation attaches to the developer, then notification windows, audit access and indemnity for late notice become priced items in access to US models. Operators with little bargaining power are the ones least able to negotiate them, which is the same fixed-cost logic the [incumbency analysis](https://theupsidejournal.com/articles/commerce-ai-board-incumbent-advantage/) applied to developers, now one level down the supply chain. This too is a labelled hypothesis: no sourced cost of these terms exists.

The verdict: the contract is the only route the text makes probable, and today's published terms do not yet supply what the clock would need.

### Route two: could the operator be a developer in its own right?

The frontier-model definition covers "a system combining multiple artificial intelligence models" that meets the performance test. The text does not say that nationality or location matters. A product that combines several models is therefore not excluded from the definition by being built in Lagos or London. Whether it meets the performance threshold is a separate question, and nothing in the text read indicates that an ordinary application built on a licensed model would.

The Section 3(f) language adds one more point of contact: the 45-day access duty is tied to introduction "into interstate or foreign commerce". Whether a non-US developer selling into the United States introduces a model into that commerce is a reading of the clause, and the text does not answer it. The verdict: this route is open, and on the text it matters only to operators building models of their own at frontier performance.

### Route three: does the critical-infrastructure limb reach beyond the United States?

Section 2 imports the definition of critical infrastructure from section 1016(e) of the USA PATRIOT Act of 2001, codified at 42 U.S.C. 5195c(e). That provision defines critical infrastructure as "systems and assets, whether physical or virtual, so vital to the United States" that their incapacity or destruction "would have a debilitating impact on security, national economic security, national public health or safety, or any combination of those matters".

The reference point is the United States. On a plain reading, a Lagos or London operator's assets fall inside the limb only if they are vital to the United States, and the limb is confined to AI used in managing industrial control systems or other operational technologies. A payments or software product that is not controlling physical processes sits outside it. This is The Upside Journal's reading of the cross-referenced definition, not a statement in the bill.

The verdict: the direct routes depend on definitions the bill leaves open or narrows, which makes the contract the working assumption and the other two the questions to monitor.

## What Does the London Signal Add?

London's position differs in kind. The UK Cabinet Office announcement of 12 October 2026, updated on 16 October, says the government is working with the National Cyber Security Centre and the AI Security Institute on how to respond to incidents involving autonomous AI agents, and that it will examine how near-miss incidents can be identified and reported. It also says the government will explore strengthening parts of the Cyber Security and Resilience Bill and the wider regime. The release describes preparation and examination; it does not describe a legal duty on any company. The Upside Crosscurrents covered the announcement as [a pilot rather than a law](https://theupsidejournal.com/crosscurrents/2026-w43/).

For a London operator the consequence is asymmetric. The nearest binding text on frontier-model incidents is not a British one, and the American one has not passed. The practical exposure in the near term is therefore commercial, through vendor terms, and not statutory.

## What Should an Operator Ask Its Vendor to Put in Writing?

Because the bill leaves "confirmed" and "provider" undefined and addresses no customers, an operator cannot rely on the statute to fill gaps. The following are The Upside Journal's recommendations for the contract, framed as questions to put to a vendor:

- What is the maximum time between the vendor confirming an incident and notifying the customer, and is it shorter than the vendor's own 30-day and 72-hour windows?
- Who decides that an incident is confirmed, on what standard, and can the customer challenge the decision?
- What information and audit rights follow a notified incident, and are independent audit reports available beyond an annual summary?
- Who bears the loss if notification is late, and is the indemnity capped?
- Does the contract reopen if a law or rule changes the vendor's reporting duties?

The verdict: a vendor that cannot answer these five questions in writing leaves the operator carrying a risk the statute does not allocate.

## What Is Unknown, and What Would Change the Conclusion?

The bill has not passed. Sens. Warner, Schatz and Kim sought unanimous consent to pass it on 29 September 2026, and Sen. Ted Cruz, who chairs the Senate Commerce Committee, objected, citing concern that the proposed Board and the Commerce Department would hold too much authority; he said he preferred a role for independent judges, according to reporting by Implicator on 30 September. That reporting also describes the bill as still unnumbered when it reached the floor, and the text published by Senator Warner's office carries a placeholder in place of a bill number. The [pillar analysis](https://theupsidejournal.com/articles/warner-ai-safety-board-frontier-model-access/) identifies the bill as S. 5576, introduced on 29 September 2026; the Section numbers quoted here are attributed to the text published by Senator Warner's office, a pre-numbering draft.

Several things remain unknown until amendment or rulemaking. The text sets out deadlines running from enactment, including a 90-day period to establish the Board and one year to establish the incident database under Section 5. It does not define "confirmed" or "provider". Standards under Section 3 are to be adopted by rule, and a rule could add detail the statute omits.

Three developments would change the conclusion of this article: an amendment that adds a territorial clause or a definition of a covered foreign person; a rule or guidance defining "provider" to include or exclude those who supply models through a commercial interface; and any Nigerian or British instrument that creates a local reporting duty. No Nigerian regulator or procurement rule was examined for this article, and that side is left for a later piece.

## Where This Article Fits

This article stops at who the duty binds and how it travels. What the bill requires sits in the [pillar analysis](https://theupsidejournal.com/articles/warner-ai-safety-board-frontier-model-access/); the clock and contract terms sit in the [incident-clock piece](https://theupsidejournal.com/articles/ai-incident-clock-vendor-contracts/); the Nigerian side is left to a forthcoming piece. The verdict: this article settles who is bound and how the duty travels; the rest sits in those pieces.

## Why This Matters for Founders and Counsel in Lagos and London

For founders, the decision is contractual and near-term: the terms to negotiate with a US model vendor are worth having whether or not the bill passes.

For general counsel, the work is to map which of the three routes could apply and to avoid two errors: assuming the duty cannot reach a foreign operator because the bill is American, and assuming it does because the vendor is. For investors, the question is whether a portfolio company's dependence on a single model vendor carries a notification risk that is priced in the contract. The [broader playbook for AI regulation](https://theupsidejournal.com/articles/ai-regulation-playbook/) places this question within the wider set of Washington rules operators outside the United States will meet first as contract terms.

## Key Takeaways

- The incident duty in Section 5(a)(3) binds developers and providers of frontier models and operators of critical infrastructure that use AI in industrial control or operational technology, and does not name the enterprise that builds on a model.
- The text read has no territorial clause and no definition of a United States person, so reach to a Lagos or London operator has to be argued from definitions and is not stated.
- The nearest route to a foreign operator is its vendor contract, where notification, confirmation, audit and indemnity terms would carry the duty; that is a hypothesis, not a mandate.
- The critical-infrastructure limb is anchored to assets vital to the United States and confined to industrial control and operational technology, which makes it a narrow route for most software and payments operators.
- The bill has not passed, and its undefined terms, "confirmed" and "provider", are the points on which a rule or amendment would change the answer.

## Conclusion

The question for a Lagos or London operator is whether the Act would reach it directly or only through its vendor's contract, and the answer sits in the bill's definitions. On the text read, the contract is the likelier route and the direct routes are open questions. The operator's task is to secure in writing what the statute does not guarantee and to watch the definitions that a rule or an amendment would settle.
