AI, Negligence and the Future Standard of Care
One of the most important legal questions emerging from artificial intelligence is not simply:
Who is liable when AI causes harm?
The deeper question is:
What should reasonable care look like before a high-risk AI system is deployed?
Traditional negligence law is fault-based. It does not usually impose liability for harm alone. Courts ask whether the risk was foreseeable, whether the harm was serious, whether reasonable precautions were available, and whether the defendant failed to meet the required standard of care.
That framework still matters.
But high-risk AI changes the timing of the inquiry.
Where AI systems are used in medicine, policing, infrastructure, finance, welfare, migration, employment, legal decision-support, or other rights-affecting domains, waiting until after harm occurs may be too late.
The harm may scale quickly.
The decision-making may be opaque.
The affected person may not understand how the decision was reached.
The institution may rely too heavily on automation.
The system may drift after deployment.
This is why I have been exploring the idea of an Accelerated Standard of Care Layer.
The argument is not that AI should automatically attract strict liability.
It is more measured than that.
The better starting point is enhanced negligence architecture: preserving fault-based liability, but raising what “reasonable precautions” require in high-risk environments.
For low-impact AI tools, ordinary governance may be enough.
But where a system affects life, liberty, bodily integrity, procedural fairness, privacy, non-discrimination, essential services, or coercive state power, reasonable care should require more.
That may include:
risk-tier classification,
foreseeable harm mapping,
bias and misuse assessment,
independent audit,
adversarial testing,
red-team simulation,
meaningful human oversight,
immutable logging,
override capability,
residual risk declaration,
post-deployment monitoring, and
recertification after material change.
The key distinction is this:
Harm alone should not equal liability.
But harm combined with ignored foreseeable risk, poor oversight, inadequate testing, or missing safeguards may support an inference that reasonable care was not taken.
In that sense, the future of AI negligence may not require abandoning traditional tort principles.
It may require operationalising them earlier.
The aim is not to turn negligence into strict liability.
The aim is to make reasonableness visible before harm becomes legally visible.
Leave a Reply