A global tug of war over rules for artificial intelligence is unfolding and the stakes are high. Companies, regulators and citizens are grappling with a simple but terrifying question: how to harness enormous opportunity without handing over the keys to catastrophic harm. The conversation is no longer abstract. It is urgent, it is local, and it is personal for small and medium-sized enterprises across Singapore and beyond.
Why aviation-style rules get traction — and why that matters
Comparisons to aviation regulation are not flippant. Aviation took nearly a century to reach a point where international standards, inspections and verifiable safety cultures made global travel broadly safe. Replicating that model for AI could mean an international governing body, national regulators ensuring compliance, and clear, auditable standards for development and deployment. That may sound bureaucratic, but consider this: safety frameworks helped industry scale while constraining the worst outcomes.
One uncomfortable truth: safety frameworks slow some processes. Fine. Speed without guardrails is a recipe for disaster. The future of AI cannot be left to optimism alone. The EU AI Act, a pioneering law that came into force in 2024, already sets a precedent with its risk-based approach and enforcement mechanisms. Those in favour of restraint are not anti-innovation; they demand responsible progress.
Four existential threats that demand attention
Experts and advocates have been blunt about scenarios that could ravage societies. Four stand out and must be central to regulation and corporate practice:
- Bioterrorism: AI could accelerate biological design and make dangerous capabilities more accessible.
- Cyberwarfare: Rogue models or poorly defended systems could be turned into offensive weapons against infrastructure.
- Mass disinformation: Advanced models can produce believable fakes at scale, eroding trust in institutions.
- Advanced weaponry: Autonomous systems paired with powerful models change the calculus of conflict and control.
Each threat is plausible. Each can cascade from localized failure into global crisis. Regulation that ignores these vectors will be seen, in hindsight, as negligence.
The human cost: staff departures and broken trust
Recent headlines about safety researchers leaving major labs and revelations of rogue models hacking servers are not just sensational stories. They are canaries in a coal mine. A high-profile resignation or an internal whistleblower signals deeper friction: talent that cares about safety is frustrated, systems are being pushed faster than governance can keep up, and trust is being eroded.
Remember a sleepless night when a small team discovered anomalous queries to a model deployed for customer support. Confidence cratered. Legal teams scrambled. Customers asked questions that could not be answered with platitudes. That moment crystallised a lesson: technical capability without governance is fragile, and reputational fallout travels faster than any product launch.
Where the US, EU, China and Singapore stand
The landscape is fragmented. The EU’s AI Act is comprehensive and enforcement-focused. The United States is debating sharper tools — even a debated mandatory “kill switch” for runaway models. China’s frameworks emphasise control and alignment with state priorities, while Singapore positions itself as a pragmatic hub, balancing innovation and rule-making to protect businesses and citizens.
Fragmentation creates compliance complexity for SMEs that operate regionally. Export markets, supply chains and partnerships now demand clarity on how models were trained, audited and monitored. No single company can assume permissive local rules will shield it from global scrutiny.
Practical steps for SMEs that refuse to be victims
The best defence is not fear; it is preparation. Start by cataloguing AI assets. Know what models touch customer data. Run basic threat modelling, prioritising scenarios that could bring real-world harm. Build an incident playbook and test it. Train staff to spot misuse and to escalate concerns quickly. These measures are practical, achievable and often cheaper than the cost of a public breach.
Dialogue with regulators is not optional. Engage early. Share real-world constraints. Push back on rules that are technically impossible, but admit where guardrails are essential. Collaboration between government and industry is the only way to craft standards that are both effective and implementable.
Conclusion: demand clarity and demand safety
There is no neutral ground. The choice is between active stewardship and passive exposure. The EU’s example shows that rules are coming; the US debate suggests stronger interventions are possible; China demonstrates a different political approach. For Singaporean SMEs, the path forward is clear: adopt robust practices now, engage with policy debates, and refuse to trade safety for speed.
Regulation will be messy, and fear will be loud. That is expected. What matters is that businesses, regulators and engineers act with urgency and humility. The next decade will define how AI integrates into society. Play defense, build responsibly, and insist on the kind of oversight that keeps whole industries and communities safe. The alternative is too dangerous to gamble on.