Skip to main content
Labs

Preparing governments for transformative AI

  • AdaptationEquipping public servants to adapt as transformative AI arrives
  • EvaluationMeasuring how AI is changing public-sector decisions and outcomes
  • GovernanceKeeping public decision-making well informed, understood and under human control as AI shapes more of it

AI is moving faster than the institutions meant to govern it

AI systems get noticeably more capable every few months. The institutions that will have to adopt, govern and live alongside them work on a much slower clock.

If transformative AI arrives on the shorter timelines many in the field now consider plausible, governments will face large calls on security, public services and accountability within the immediate next few years. We think preparation has to start now, with the people inside government who will make those calls.

How much work can AI do on its own?

The length of task leading AI models can complete on their own, in hours a skilled professional would take. It has roughly doubled every four months since 2023.

Length of task leading AI models can complete on their own, by release date
DateModelTask length (skilled human work)
March 2023GPT-44 minutes
October 202420 minutes
December 202438 minutes
February 20251 hour
April 2025o32 hours
August 2025GPT-53.6 hours
November 2025Claude Opus 4.55.3 hours
December 20256.6 hours
February 2026Claude Opus 4.612 hours
April 2026Mythos Preview (early result)17.4 hours
Data: METR, Time Horizon 1.0 and 1.1, measured on technical tasks at 50% reliability. The latest point is an early result.

Who we work with

  • Public servants

    Apolitical's community of 500,000+ public servants: people preparing for more capable AI, or already using it and wanting an independent view of what's working.

  • Researchers

    Working on AI safety, evaluation or governance and looking for a route into real government settings.

  • Partner organisations

    Institutes, civil society groups and companies working on safe and beneficial AI who want to understand how it plays out inside government.

Labs is the transformative AI team at Apolitical.

We build on Apolitical's global community of public servants, testing ideas with the people who will use them and learning quickly from what they tell us.

Working on AI in government, inside it or alongside it? We'd like to collaborate: labs@apolitical.co