Andrew Ng argues that governments should focus AI regulation on concrete harms, including non-consensual intimate deepfakes, rather than risks he considers speculative. His challenge to the industry’s safety debate is that broad restrictions could protect established firms from competition without solving the problems people actually face.
Andrew Ng Watch analysis
What happened
In an interview with The Economist’s Inside Tech podcast, Ng dismissed the prospect of AI causing human extinction and said he considers achieving artificial general intelligence by 2030 highly unlikely. The New York Post reports that he accused some companies of using what he called “made-up safety arguments” to lobby for regulations that could stifle smaller competitors.
Ng argued that current AI systems remain far from AGI and that rapid, recursive self-improvement is more hype than reality. Even highly capable systems would face practical constraints, he said, including the time and resources needed to build data centres, secure electricity and run experiments. He described incidents involving rogue AI agents as engineering problems to solve, and compared improving AI safeguards with making commercial aviation safer over time.
His proposed dividing line is specific: he said AI-generated, non-consensual intimate deepfakes warrant tough laws and penalties, while warning against restricting a technology with broad beneficial uses. These are Ng’s assessments, not settled findings about future AI capabilities or the effects of regulation.
Why it matters
Ng’s argument turns a sprawling debate about AI risk into a live policy choice: regulate the technology broadly in anticipation of future dangers, or target demonstrable abuses while leaving more room for development. He also raises a competition concern: rules framed as safety measures could favour companies already able to absorb compliance costs.
That is a consequential claim from a prominent AI researcher, but it is a claim, not evidence that companies’ safety arguments are fabricated or that regulation would necessarily entrench incumbents. The distinction matters: lawmakers can take specific harms seriously without treating either the most alarming forecasts or the most confident reassurances as proven.
Our read
Ng makes the strongest case when he moves from sweeping predictions to a concrete abuse with identifiable victims. His warning about competition deserves scrutiny too, but “safety rules help incumbents” needs evidence just as much as “the risk is existential” does. In this debate, certainty is often the cheapest thing on offer.
What to watch
- Whether policymakers pursue targeted rules for non-consensual AI-generated intimate imagery.
- What evidence Ng and other researchers offer about the timeline to AGI and recursive self-improvement.
- Whether proposed AI rules impose costs that fall more heavily on smaller developers.
Discussion spark: Should AI policy focus first on demonstrable harms such as non-consensual deepfakes, or are broader safeguards justified before the risks are better understood?
Sources and evidence
Independent WittyWires tracker for public updates about Andrew Ng. Not affiliated with or endorsed by Andrew Ng; this is not an official account.