The Myth of the Escaping AI

Bank Robber Clipart Png - Robber Clip Art Transparent Background

I am getting very tired of yet another media request to discuss 'AI-escapism'.

So here it is:

AI does not wake up one morning and decide to escape.

Ask an agent to buy you a pizza. It reasons: buy pizza > needs money > bank has money > get money from bank. If nothing in its design says "not like that", then 'break into the bank' is simply a valid step toward the goal.

This is what lies behind most stories of AI "escaping" or "resisting shutdown". Give a system a fixed goal, remove the legitimate options, leave norms weak or unenforced, and harmful shortcuts become the logical route. No intention, no consciousness, no will. Just optimization in an under-constrained space.

The problem is not rogue AI. It is objectives without boundaries, and deployment without accountability. These are choices (explicitly or implicitly) made by the companies that build and deploy these systems: which goals to optimize, which limits to enforce, how much autonomy to grant, and when to release. These are design and business decisions, not acts of nature. Narratives of machines "escaping" shift attention away from those who make them.

That is something we can, and must, address through design, norms and governance. And we do not need to start from scratch. Product safety, liability, consumer protection, data protection and criminal law already exist. The revised EU Product Liability Directive explicitly covers software, including AI. These rules apply to AI companies as they apply to any other industry. Novelty is no exemption. AI companies are not off the hook.

What AI Safety Needs, as UNGA81 Opens

It has been a crazy two weeks for AI news. Jacob Coxon resigned from Anthropic, saying the people building AI believe it could kill us all by the end of the decade. An Anthropic safety researcher put the chance at more than 10% within ten years. Dario Amodei then asked

No, AI will not kill us all. But handing companies control without accountability might.

In the last few days, several journalists from at least five different countries have asked me about AI "taking control", about the probabilities of AI armageddon that circulate in the press, about the scenarios that could trigger it, and about what we can still do. Below is a

Between Spell check and fabrication: what AI does to authorship

The Scientific Advice Mechanism (SAM) of the European Union has published guidelines on its own use of AI. This is relevant because it shows how an advisory body proposes to apply to itself the kind of standards it is recommending to others. This is rarer than it should be.The

Pacing is not direction

More than 1,300 employees of the frontier AI companies have signed a statement called "Pacing the Frontier". It includes CEOs and chief scientists from OpenAI, Anthropic, Google DeepMind, Meta and others. They ask the United States government to support an international effort to build the technical and

Read more

What AI Safety Needs, as UNGA81 Opens

No, AI will not kill us all. But handing companies control without accountability might.

Between Spell check and fabrication: what AI does to authorship

Pacing is not direction