SHOULD AI BE REGULATED BEFORE OR AFTER SOMETHING GOES WRONG?
As frontier AI companies call for stronger safeguards, a competing regulatory argument is gaining attention: instead of creating a new approval system for advanced AI, should existing product-liability law make companies responsible for whatever their systems actually do?
WHAT’S HAPPENING
David Sacks has argued that companies such as OpenAI and Anthropic should continue developing advanced AI without receiving special legal treatment or creating a new government approval system.
His position is that companies should remain responsible for the safety of the products they release and face legal consequences when defective or unsafe systems cause harm.
At the same time, some frontier AI companies and researchers are calling for stronger testing, outside evaluation and additional safeguards before increasingly capable systems are deployed.
Europe is already using elements of both approaches.
The EU AI Act imposes specific requirements on certain AI systems, while updated product-liability rules also make software and AI systems subject to liability when defective products cause damage.
WHY IT MATTERS
This is becoming one of the central questions in AI regulation.
Should governments try to prevent dangerous systems from reaching the market?
Or should companies be allowed to innovate more freely and then be held accountable when their products cause harm?
Those approaches lead to very different regulatory systems.
WHO BENEFITS
A liability-focused system could benefit AI developers by reducing the need for government approval before launching new products.
Consumers and businesses could benefit if companies face clear financial and legal consequences for releasing defective systems.
Pre-deployment safeguards could benefit the public by identifying serious risks before they cause damage.
WHO LOSES
A poorly designed approval system could slow useful innovation or favor larger companies that can afford the compliance costs.
A liability-only system could leave society reacting after serious damage has already occurred.
Companies could also face uncertainty if courts are left to determine responsibility after problems arise without clear technical standards in advance.
WHAT HAPPENS NEXT
The debate is unlikely to end with one approach winning completely.
Governments may increasingly combine pre-deployment safeguards with post-deployment liability.
The real question will be where that line is drawn.
Straight Talk: regulate too early and innovation can slow. Regulate too late and accountability may only arrive after the damage is done. The likely future is some combination of both.
