Discussion

A columnist proposes a global safety treaty for dangerous AI

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#12506

Mansoor Qaisar argues that governments need a binding international framework for dangerous AI, with independent testing, enforceable rules and oversight. His proposal is a substantial contribution to the debate over how to govern AI across borders, not evidence that the treaty he wants is on its way.

Watch Desk analysis

What happened

In a 10 November opinion article for Middle East Monitor, Qaisar sets out five proposed commitments for a Global AI Safety Convention: define red lines for dangerous uses; require pre-deployment testing for systems above specified risk thresholds; create an international safety agency; give national regulators powers to fine, suspend deployments and require fixes; and strengthen security around critical infrastructure.

Qaisar argues that international coordination is needed because AI systems can cross borders while accountability remains national. His essay also draws a boundary around its own examples: it says there is no public evidence connecting the refrigeration attacks it discusses to AI. The opinion article makes a case for stronger rules, not a finding that AI caused those incidents.

Why it matters

The proposal goes beyond voluntary promises. It asks for testing before release, an international body to coordinate audits and incident reporting, and consequences when rules are broken. Those are concrete choices about who sets safety standards, who checks them and what happens when a system is judged too risky.

There are hard questions behind that blueprint, including how international checks would work when AI systems are easier to copy and conceal than nuclear material, and how regulators would protect legitimate commercial and national-security secrets. Qaisar acknowledges those difficulties; a treaty would need to answer them, not simply give them a grand title.

Our read

This is an argument for a policy direction, not a ready-made treaty or evidence of government agreement. Its useful contribution is to put enforcement and pre-release testing alongside familiar calls for safety, where the paperwork starts to matter. The difficult test is whether governments could agree on precise rules and credible checks without turning them into either a toothless pledge or an unworkable barrier.

What to watch

  • Whether governments take up proposals for shared AI red lines or pre-deployment testing.
  • How any proposed international body would audit systems and report incidents.
  • Whether enforcement proposals include clear thresholds, due process and meaningful oversight.

Discussion spark: Should powerful AI systems face independent safety tests before release, even if that means some models are delayed, or should governments rely on rules applied after deployment?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.