AI does not need fear - it needs guardrails

Artificial intelligence is often treated as an autonomous threat. But the real question is whether governments and corporations can be made answerable for the way they introduce and use powerful technology, writes David Higginbottom.

Artificial intelligence has entered public debate like as an existential threat. It will take jobs, undermine education, flood the internet with lies, intensify surveillance and make people irrelevant – even threaten our existence. Some fears are well founded - AI can already be used to deceive, discriminate, monitor and concentrate power.

AI does not decide to replace a worker, deny a person a service, identify a target, imitate a voice or bury a fact beneath generated noise.

Governments, corporations, managers, designers and investors make those decisions within systems of law, ownership, procurement, regulation and public oversight – and if these guardrails do not exist, in the gaps where those systems should be.

The question about AI is not whether it is good or bad. It is whether we have democratic guardrails to control its use.

The comparison with defence technology helps expose the issue. Governments spend enormous sums on submarines, aircraft, missiles, intelligence systems and autonomous military capabilities.

These projects are described in the language of security, deterrence, jobs, alliances and national capability. A new AI tool, by contrast, can provoke instant warnings about social collapse.

AI merits attention because it reaches people directly: at work, in schools, on phones, in welfare systems, policing, and surveillance in stores.

However, an AI system is presented as disruption; a military system as protection.

In both cases, the question is the same: who has decided that it should be used, for what purpose, under what limits and with what public authority?

A submarine does not become harmless because it is familiar, expensive or funded by government. Nor does AI become dangerous simply because it is new. Both acquire their social meaning through the purposes they serve and the controls - or lack of controls, in their development and use.

The risks are not identical. A chatbot that produces false information is not a missile system, and a flawed hiring tool is not a decision to go to war. Major conflict between nuclear-armed states carries the risk of escalation that could destroy civilisation – and that risk is treated as background noise.

Global warming - or climate reversal - should dominate this debate.

It is the present, systemic threat to the stable conditions on which human security depends.

As I argued in It’s not climate change - it’s climate reversal, we are not simply living through another natural variation. In little more than a century, we have begun reversing the geological process that removed carbon from the atmosphere. That multi-million-year sequestration process created the stable climate in which civilisation developed — and on which our future depends.

A nation cannot claim to be secure if it prepares for military contingencies while failing to address a crisis already intensifying heat, drought, fires, floods, displacement, food stress and conflict.

Defence is part of that crisis, not outside it. Military forces rely on fuel-intensive aircraft, ships, vehicles, bases, supply chains and weapons production. Their emissions are substantial but poorly disclosed, as the Conflict and Environment Observatory has documented.

War destroys infrastructure and leaves contaminated water and soil, damaged ecosystems, fires, toxic debris and the long afterlife of unexploded weapons. It also kills and displaces people on a vast scale. SIPRI has warned that rising military expenditure can divert resources from sustainable development; in a climate emergency, that directly competes with investment in emissions reduction, clean energy, adaptation and protection for communities already exposed to climate damage.

This does not mean abandoning every defence capability. It means security cannot be reduced to military strength. A policy that adds to environmental breakdown while claiming to keep people safe, has not passed the public-interest test.

We should stop treating technology as destiny. Societies choose whether to fund a system, who owns it, what data it uses, where it is deployed, who profits, what harms are acceptable and whether the public has a right to know how it works. When governments say they must acquire a capability because the strategic environment demands it, or companies say they must deploy AI because competitors are doing so, they present political choices as natural laws - a technological race is not a democratic argument.

National-security language also can shut down questions. Deterrence, interoperability and strategic necessity may describe genuine concerns, but security claims should make scrutiny more rigorous, not less. Before committing to a major military capability, a government should explain its purpose, independent evidence, legal basis, alternatives, cost, climate and environmental effects, risks and conditions for review. Parliament and the public should have more than a ceremonial role after key decisions have been made.

The same principle applies to AI. Before an automated system is embedded in a public service, workplace, school, police force or critical infrastructure, people should be able to ask: What problem is it solving? Is it necessary? What evidence shows it works? Who is responsible when it fails? Can an affected person challenge it? Can it be suspended if it causes harm?

These are not technical questions. They are questions of democracy.

The guardrails are clear. First, there must be a public purpose: technology should not be introduced merely because it is fashionable, well marketed or justified as keeping pace in a race. Second, there must be firm limits: systems that make unchallengeable decisions about liberty, livelihood or essential services should not be accepted merely because they are efficient; military advantage cannot remove legal, environmental or moral limits on force.

Third, there must be transparency. Necessary secrecy and commercial confidentiality cannot become blanket excuses for hiding contracts, conflicts of interest, testing results, emissions, failures or operating rules. If public money is used or public rights are affected, the public interest must be visible.

Finally, there must be accountability and continuing control. Responsibility cannot disappear into an algorithm, a contractor or a security classification. People need a right to an explanation, human review and remedy. Whistleblowers, journalists and independent regulators must be able to expose wrongdoing. Public institutions must retain the authority to audit, restrain, redesign or withdraw systems that fail the public-interest test.

The real danger is not artificial intelligence alone. It is intelligence - artificial or human - placed in systems that are secret, unaccountable and insulated from challenge.

We do not need to decide whether to fear AI more than submarines, missiles or surveillance systems. We need a consistent standard for every high-consequence technology: who decides, who benefits, who bears the risk, who can challenge the decision, what ecological damage results, and what happens when it fails.

Technology will continue to change. The question is whether democratic control changes with it - or whether we allow the language of innovation and security to make public power less visible, less accountable and harder to reclaim.