Originally published August 29, 2025 · Updated October 17, 2026
Also published by RegulatingAI, a Diakon Partners client.
Updated October 17, 2026: Revised policy references and historical claims, with practical priorities for policymakers and organizational leaders. Primary sources are linked throughout.
Start With the Outcome the Rules Must Deliver
Every disruptive technology asks policymakers to make decisions before every consequence is understood. Leaders must protect the public, create room for useful innovation, and give organizations enough clarity to invest.
Experience across nuclear energy, biotechnology, finance, transportation, and medicine offers a useful starting point. These industries do not provide a ready-made rulebook for artificial intelligence. They do show why evidence, clear responsibility, public participation, and the ability to correct failures belong in the design of governance from the beginning.
Three Priorities for Leaders
- Define the risk and the desired result: Identify who could be affected and what evidence would demonstrate acceptable performance.
- Make responsibility clear: Establish who tests, approves, monitors, and responds when a system fails.
- Build in review: Revisit requirements as experience, technology, and the law change.
What Other Technologies Teach Us
Nuclear Energy: Oversight Must Reach the Operating System
The Nuclear Regulatory Commission’s Three Mile Island knowledge management digest describes changes following the 1979 accident in training, emergency preparedness, equipment, and regulatory oversight. The lesson extends beyond any single technical safeguard: people, procedures, equipment, and oversight interact.
For AI, a sound review should examine the setting in which a model is used. Who interprets an output? What happens when data is incomplete? Can an employee stop or override the process? A technically capable model can still perform poorly inside an operation that lacks those controls.
Biotechnology: Early Deliberation Can Shape Practical Safeguards
The 1975 Asilomar conference brought scientific concerns about recombinant DNA research into a structured discussion of risks and safeguards. Its summary statement remains an important historical example of researchers addressing the conditions under which work should proceed.
It does not establish that early dialogue prevents all backlash or resolves every ethical question. It does suggest a useful approach for AI: involve the people developing, deploying, overseeing, and experiencing the technology before operational choices become difficult to reverse. Include workers, customers, affected communities, and smaller organizations alongside technical and policy specialists.
Automated Vehicles: Separate the Different Layers of Authority
The U.S. Department of Transportation describes automated-vehicle responsibilities across vehicle safety, commercial operations, highways, and other parts of the transportation system. Oversight involves several agencies and levels of government.
The governance lesson is to identify which authority addresses which risk. Calling every difference “fragmentation” can obscure legitimate responsibilities; leaving overlapping requirements unexplained can make compliance difficult. AI policy needs a clear account of provider duties, deployment responsibilities, sector rules, and the roles of federal and state authorities.
Digital Finance: Legislation and Implementation Are Different Milestones
The Guiding and Establishing National Innovation for U.S. Stablecoins Act, or GENIUS Act, became Public Law 119-27 on July 18, 2025. It provides a concrete example of Congress establishing a statutory framework for a defined financial technology.
Enactment is one milestone. Effective dates, agency rules, supervisory capacity, and the activity covered by a law determine how the framework operates. For AI, policymakers and businesses should distinguish a proposal, an enacted statute, an implementing rule, and an obligation that already applies. Announcing a framework does not complete its implementation.
Aviation: Shared Standards Need Continuing Oversight
ICAO’s Universal Safety Oversight Audit Programme assesses states’ safety-oversight systems and implementation of international standards and recommended practices. The framework connects common expectations with assessment and follow-up.
AI governance can draw on that emphasis on continuing oversight without assuming that the same institutional structure fits every application. Comparable testing methods and incident information can help organizations learn across borders. They do not remove the need to identify the law and regulator responsible for a particular deployment.
Medicines: Evidence Should Match the Consequences of Failure
The FDA’s drug-development process includes research, clinical evaluation, regulatory review, and monitoring after approval. Evidence is gathered before use and continues to be assessed afterward.
The AI lesson is proportionate evaluation. A tool drafting an internal meeting summary and a system influencing a consequential decision warrant different levels of scrutiny. Define the intended use, test foreseeable failure modes, document limitations, and decide how performance will be monitored after deployment.
From Industry Experience to AI Practice
| Industry | Governance lesson | Apply it to AI |
|---|---|---|
| Nuclear energy | Review people, equipment, and procedures together. | Test the workflow, escalation process, and human override. |
| Biotechnology | Address risks and safeguards before choices become difficult to reverse. | Involve developers, users, workers, and affected communities early. |
| Automated vehicles | Clarify responsibilities across authorities. | Map provider, deployer, sector, and jurisdictional duties. |
| Digital finance | Distinguish enactment from implementation. | Track legal status, effective dates, rules, and oversight capacity. |
| Aviation | Connect shared standards with continuing oversight. | Use comparable tests, incident learning, and follow-up reviews. |
| Medicines | Match evidence to the consequences of failure. | Set testing and monitoring requirements for the intended use. |
What Has Changed Since the Original Article
Policy developments reinforce the need to check the scope and timing of each requirement.
- European Union: The Commission reports that the AI Omnibus entered into force on July 27, 2026. Its updated timetable places Annex III high-risk requirements at December 2, 2027 and high-risk systems embedded in regulated Annex I products at August 2, 2028. Other provisions apply on different dates. The official implementation timeline should be read with the applicable text and transition provisions.
- United States: The White House’s March 2026 national AI legislative framework set out objectives and called for congressional action. That announcement itself did not enact those objectives into law.
- Environmental-policy example: EPA finalized its rescission of the 2009 greenhouse-gas Endangerment Finding and repeal of motor-vehicle GHG standards in February 2026. The original article described a 2025 proposal. This example concerns the specified motor-vehicle action; it should not be read as a statement that all environmental requirements disappeared or as a determination of subsequent litigation.
For a business or association, the practical response is a record of the relevant jurisdiction, activity, legal instrument, effective date, and implementation responsibility. Verify any later amendments or court orders before changing a compliance process.
Voluntary Frameworks and Enforceable Rules Have Different Roles
Voluntary guidance can support useful work before or alongside legislation. The NIST AI Risk Management Framework is intended for voluntary use in incorporating trustworthiness into AI design, development, use, and evaluation. NIST also identifies an ongoing revision of AI RMF 1.0.
A framework can help organize testing, governance, and vendor questions. It does not certify compliance or replace applicable legal duties. Conversely, a statute needs the people, procedures, evidence, and resources required to implement it. Effective governance should make those relationships understandable.
Turn Principles Into Decisions People Can Follow
Organizations can begin with a small set of concrete questions:
- What is the system allowed to do? Define its purpose and the decisions that remain with a person.
- What evidence supports deployment? Test representative conditions and specify the performance threshold.
- Who is accountable? Assign an owner for approval, monitoring, incidents, and changes.
- How can affected people raise concerns? Provide a workable channel for questions, correction, and escalation.
- When will the decision be revisited? Set review triggers for material changes in the system, its use, or the governing requirements.
These are management priorities, not a claim that every AI application faces an identical legal process. They help organizations connect broad principles with daily decisions and provide policymakers with evidence about what works.
Balance Requires Follow-Through
The recurring lesson from other technologies is that governance must remain connected to the activity it governs. Clear expectations help organizations plan. Proportionate evidence helps test whether safeguards work. Ongoing review makes it possible to correct problems without waiting for a crisis.
AI policy will be judged by those results: useful innovation, credible accountability, and protections that operate in practice. Building the capacity to deliver them is a shared responsibility for policymakers, industry, researchers, and the communities affected.
Contact Diakon Partners to discuss how technology policy developments affect your organization’s strategy and stakeholder engagement.
Sources and Revision Note
This October 17, 2026 revision updates the EU AI timetable and the EPA example, clarifies the status of legislative announcements, and removes unsupported market, public-opinion, telecommunications, and aviation statistics. Historical analogies are presented as lessons for governance rather than proof that one policy caused an industry outcome. The original article was also published by RegulatingAI; this revision is the current Diakon Partners website edition.