ControlAI publishes 'A Narrow Path' policy plan to prevent superintelligence development

ControlAI, a nonprofit founded by Andrea Miotti to campaign against the development of artificial superintelligence, published “A Narrow Path,” a policy document proposing a phased route to keeping advanced AI under human control.

The plan set out three phases. “Safety” called for immediate national legislation and new institutions to prevent anyone from developing uncontrolled superintelligence for roughly two decades. “Stability” proposed international oversight arrangements robust enough to survive geopolitical competition and prevent rogue development by states or other actors. “Flourishing,” contingent on the first two succeeding, envisaged safely developing transformative AI once control was assured. The authors said they had found no other plan that tried to address AI extinction risk this comprehensively, which was their reason for writing one; it drew comparisons to nuclear non-proliferation regimes.

The document received limited attention on release but became the basis for a sustained ControlAI lobbying campaign in the UK Parliament, briefing cross-party MPs and peers over the following two years and backing a 2026 amendment giving government powers to shut down datacentres judged a critical security threat. Critics, including some AI policy researchers, argued the plan’s twenty-year prohibition on frontier development was politically unrealistic and underspecified on enforcement mechanisms; supporters treated it as the first attempt at a complete international governance blueprint rather than a single-country bill.