(Sept 17): For much of the past three years, the politics of artificial intelligence (AI) has revolved around relatively familiar questions. Will AI take jobs? Who owns the copyright to the material on which models are trained? Can companies protect personal data? And who should be responsible when an algorithm causes harm?

Britain’s latest AI debate suggests we may be entering a very different phase. More than 70 members of Parliament and peers have urged Prime Minister Andy Burnham to support legislation prohibiting the development of artificial superintelligence (ASI) and pursue an international agreement preventing its creation.

The proposal is unlikely to become government policy immediately. ASI remains hypothetical, its definition contested, and Britain continues to see advanced AI as an important source of economic growth and strategic advantage.

But focusing on whether Westminster actually bans superintelligence misses the more important development.

The politics of AI safety is moving from technology policy into national security. Once that happens, the threshold for government intervention changes. Policymakers become more willing to impose restrictions despite economic costs. 

Voluntary commitments become less persuasive. Companies accustomed to dealing with technology ministries and regulators suddenly encounter defence establishments, security agencies and heads of government.

Telecommunications infrastructure and semiconductor supply chains have already undergone versions of this transition. AI may be travelling along the same path.

What makes the latest debate particularly significant is that calls for restraint can no longer easily be dismissed as coming from people outside the technology industry.

Anthropic chief executive officer Dario Amodei has argued that AI companies should deliberately pace the rate at which capabilities advance, giving safety research and safeguards time to catch up.

His position is not simply to stop AI development. Amodei continues to argue that AI could deliver enormous benefits. His concern is that capabilities may now be advancing faster than our ability to understand and control them, particularly as AI itself becomes increasingly useful in developing subsequent generations of AI. But something more striking has now happened.

OpenAI CEO Sam Altman has publicly backed Amodei’s argument that the industry needs to “pace the frontier”, saying it has been a primary subject of discussion inside OpenAI. He has also committed OpenAI to Amodei’s proposal to give independent evaluators employee-like access to assess safety practices.

Elon Musk, whose xAI competes directly with both companies, offered an even more succinct endorsement: “Dario is right.”

The significance lies less in the individual statements than in who is making them.

These are fierce commercial competitors with very different views about AI and its governance. Yet leaders of three major frontier AI companies are now publicly acknowledging that there may be circumstances in which capability development should slow.

OpenAI has gone further. It is advocating mandatory, capability-based national AI safety regulation, independent safety assessments and international standards for determining when development should slow or stop.

That represents an important change in the regulatory debate. The question may no longer be simply whether governments should accelerate or constrain AI. It could become how fast the frontier should move, and what safeguards must accompany each increase in capability.

This could have profound consequences for business. Amodei has proposed giving independent external evaluators ongoing access to parts of Anthropic’s operations, comparing the concept with regulatory supervisors embedded within financial institutions. OpenAI has now said it will do the same.

The analogy should attract policymakers’ attention. Financial regulation did not develop on the assumption that banks could simply declare themselves safe. Independent supervision, stress testing, capital requirements and disclosure became embedded in the system.

Something similar could eventually emerge around frontier AI. Saying that a company takes safety seriously may no longer be sufficient. Governments may demand that companies demonstrate it through independent testing, incident reporting and measurable thresholds beyond which additional safeguards become mandatory.

AI governance could therefore be moving from promises towards verification.

The Burnham government consequently faces a difficult balancing act. It wants Britain to be a serious AI power, requiring investment, infrastructure, talent and companies willing to develop increasingly capable models.

Yet it must simultaneously convince voters that those technologies will not create unacceptable risks.

There is no simple national solution because AI development is inseparable from geopolitical competition. If democratic countries slow their programmes while competitors do not, restraint could create a national-security vulnerability.

That makes international coordination increasingly important.

Britain will host the Group of 20 in 2027. Rather than an improbable global prohibition on superintelligence, governments may find more practical territory for cooperation: prohibiting narrowly defined dangerous applications, developing common testing standards for cyber and biological risks, or establishing internationally recognised capability thresholds.

Once policymakers start discussing AI in the conceptual language of arms control rather than digital regulation, politics have fundamentally changed.

For Singapore, this matters.

Singapore has deliberately pursued a pragmatic model of AI governance: encourage adoption and innovation while developing frameworks for testing, accountability and risk management.

Its experience with AI assurance, combined with its position as a trusted and technologically sophisticated economy, could give it a useful role in shaping the standards and verification mechanisms that a more internationally coordinated system would require.

For business, this is ultimately the lesson. AI companies can no longer treat regulation as a compliance exercise conducted after technology strategy has been decided. Boards need to understand how political perceptions of their technology are changing. 

Government affairs teams need relationships extending beyond technology ministries. Frontier developers may increasingly need to accommodate independent scrutiny that once would have seemed commercially intrusive.

Companies using advanced AI should pay attention too. If regulation becomes capability-based, obligations may increasingly follow what an AI system can do rather than the industry in which it is deployed.

The immediate proposal to ban artificial superintelligence may succeed or disappear. But something more consequential has already happened.

Politicians are contemplating prohibiting the most advanced forms of AI. Leaders of competing frontier laboratories are openly discussing whether development sometimes needs to slow. OpenAI is advocating mandatory capability-based safety regulations. And proposals involving independent monitors, regulatory thresholds and coordinated restraints on development are moving towards the mainstream.

The important question therefore no longer be whether greater regulation is coming.

It is whether safety can keep pace with capability — or whether governments and the companies building the technology will eventually decide that capability itself must slow down.

At Penta, Shawn Balakrishnan is the partner overseeing Asia-Pacific — his work focuses on helping organisations understand stakeholder sentiment, emerging risks, and the broader narratives shaping public affairs and corporate reputation.