Regulation is not a brake but a matter of direction.
In the economy ahead, regulation is not a footnote. It is the arena in which it is decided who captures AI gains, who is liable, and which values are embedded.
Where do things stand today, and where do they break?
The EU AI Act (in force since August 2024) is the world's first comprehensive AI framework law, with high-risk categories, transparency requirements, and liability rules for providers and deployers. The United States relies on executive orders and sectoral regulation (NIST AI RMF). China combines strict control with active promotion: algorithm regulation (2022), generative AI rules (2023). Companies face compliance costs, but also gain competitive advantages. Those who build in regulatory compliance early gain market access.
AI systems act with increasing autonomy: in medicine, justice, and financial advice. Who is liable when an algorithm discriminates or causes harm? Existing product liability law does not fit learning systems. The timing dilemma: regulation too slow → loss of trust. Too aggressive → innovation moves elsewhere.
Paths out of the crisis
The stars show how realistic a path is: 1 star = barely feasible, 5 stars = highly feasible.
Strict platform liability
AI providers are liable for damage caused by their systems, comparable to product liability.
+ Clear accountability, safety incentives.
Stifling innovation.
Trade-off: Safety vs. speed.
Feasibility: The EU is taking this path for high-risk AI.
Regulatory sandboxes
Controlled testing environments for AI under simplified rules.
+ Promotes innovation and learning from real-world deployment.
May become a permanent state.
Trade-off: Flexibility vs. commitment.
Feasibility: The UK, Singapore and EU member states apply this approach.
International AI Governance
A global AI equivalent of the IAEA: a multilateral supervisory authority.
+ Prevents regulatory arbitrage.
Highly complex politically.
Trade-off: Effectiveness vs. enforceability.
Feasibility: First steps: Bletchley Park Declaration 2023.
Principles-Based Regulation
Binding principles without prescriptive technical rules.
+ Technology-neutral, adaptable.
Difficult to enforce.
Trade-off: Future-proofing vs. clarity.
Feasibility: The OECD AI Principles follow this approach.
Trigger-based phases
Trigger: Divergent regulatory regimes without coordination
Compliance costs rise. Regulatory arbitrage becomes possible. Early liability cases set precedents.
Trigger: First international AI treaties / OECD standards
The regulatory landscape consolidates. Compliance becomes a competitive advantage. A liability framework is established.
Trigger: AI governance as part of multilateral trade agreements
Regulatory-compliant AI becomes a market-access requirement. International standards prevail.
Real-world pilot programmes
Where the ideas behind this pillar are already being tested. "Learn more" opens the explanation without leaving the page.
What we are watching
Providers in high-risk categories that are compliant by 2025: a trigger for the Phase 2 market consolidation.
Track in the index →The number of countries without AI-specific liability rules is falling: international convergence is beginning.
Documented litigation with AI relevance (2024). Beyond roughly 500 cases, robust case law begins to emerge.
Our assessment
At Globalance, we regard the liability question as the point on which AI regulation will be decided, not transparency obligations. Transparency can be satisfied without anything changing. Liability changes incentives immediately: whoever stands behind the harm caused by a system builds it differently. That is precisely what remains unresolved. Product liability is designed for objects that do not continue to develop after they are sold.
A viable approach combines principles-based regulation and Regulatory sandboxesPrinciples outlast technology cycles; detailed technical rules do not. A rule that names model architectures is out of date by the time it comes into force. The sandbox provides what every regulation needs and rarely has: experience before commitment. Strict platform liability belongs alongside it as a fallback, for cases where no one else is liable.
We advise against an international supervisory authority modelled on the International Atomic Energy Agency. Not because the idea is wrong, but because it founders on the timeline. Such institutions take decades to establish. By then, it will already have been decided who sets the standards. Waiting means surrendering influence and receiving nothing in return.
From a Swiss perspective The situation is uncomfortable: market access is governed by European law, while legislation follows Swiss timing. Anyone building here effectively complies with two sets of rules and has a hand in writing neither. A domestic law on algorithmic accountability would change little; an early, binding contribution to European standardisation would change more.
What this means for investors: First, liability for learning systems will be settled, either in law or before courts. Providers who document deployment and accountability clearly today carry the smaller risk. Second, AI in medicine, credit decisions and public administration is likely to be deployed first in supervised test environments. That is where the first viable business models will emerge. Third, those who participate in European standardisation bodies shape the rules by which products will later be sold. Fourth, auditability is becoming a quality criterion: models that can be tested transparently, without exposing their internal workings, have an advantage with regulated clients.
Why now: Regulation is written in few places and over few years. After that it becomes settled law, and settled law requires far more effort to change than rules still being formed. For investors this means that compliance is a quality signal in this phase. Companies that treat it early as a design task rather than a cost centre will retain access to institutional markets when that access becomes harder to secure.
The strongest objection concerns the distribution of compliance costs. When regulatory conformity becomes an advantage, it becomes an advantage for the large: corporations bear the fixed costs of regulation more easily than smaller providers, and a rule designed to prevent abuse thereby entrenches the market position of those best placed to meet it. There is also the question of location: development moves to where less is scrutinised. The claim that an early mover sets a global standard rather than merely bearing a competitive disadvantage has not been demonstrated. The opposite is well documented in other technology fields.
What we do not yet know
- How do we prevent regulation from shifting innovation to non-EU jurisdictions?
- Who should audit AI systems: government authorities, private certifiers, or international bodies?
- How can liability be allocated for adaptive learning systems?
- Is a global AI equivalent to the IAEA realistic?
