A short history of AI open letters
The modern AI safety conversation has been shaped, more than most people realize, by a small set of open letters. The 2015 Research Priorities for Robust and Beneficial Artificial Intelligence letter, hosted by the Future of Life Institute, gathered 11,251 signatures including Stuart Russell, Demis Hassabis, Geoffrey Hinton, Yann LeCun, Peter Norvig, Stephen Hawking, and Elon Musk. It set the agenda for the next decade of safety research by articulating concrete priorities around robustness, transparency, and value alignment.
The 2023 Pause Giant AI Experiments letter asked labs to pause training of systems more powerful than GPT-4 for at least six months. It became a cultural flashpoint but produced no policy effect. Labs did not pause. The letter is still cited regularly, mostly as a marker of the moment when the AI safety conversation went mainstream.
The October 2023 Encode Justice and FLI letter, AI Licensing for a Better Future, asked for three specific US policy actions and got a mixed hearing. Parts of its agenda have been adopted in subsequent legislation; parts have not.
The two letters that actually matter right now
The 2025 Statement on Superintelligence called for a prohibition on the development of superintelligence, not lifted before there is broad scientific consensus that it will be done safely and strong public buy-in. Initial signatories included AI pioneers Yoshua Bengio and Geoffrey Hinton, along with Steve Bannon, Glenn Beck, Obama's National Security Advisor Susan Rice, Steve Wozniak, Richard Branson, five Nobel Laureates, former Irish President Mary Robinson, and actors Stephen Fry and Joseph Gordon-Levitt. The letter has roughly 31,810 signatures and a closed comment period.
The 2026 Pro-Human AI Declaration, with over 134,015 signatories, frames the choice as a fork: replace humans in ever more roles and concentrate power in unaccountable institutions, or build AI that amplifies human potential, protects dignity and liberty, and strengthens communities. The signatories do not call for a pause. They call for a direction.
What the pattern tells us
Three things stand out across the letters. First, the ones that specify concrete policy actions have a better policy afterlife than the ones that ask for pauses. Second, the ones that frame the choice as a fork between two directions get more signatures than the ones that frame it as a single demand. Third, the credibility of the signer set matters more than the size of the signer set once the count passes a few thousand.
The 2015 letter worked because it articulated research priorities that technical people could engage with. The 2023 pause letter failed because it asked labs to do something that was not in their interest. The 2025 superintelligence letter worked because it drew a coalition that no single lab could dismiss. The 2026 Pro-Human Declaration worked because it offered a positive vision rather than a prohibition.
What to watch next
The interesting question is whether any of these letters have actually moved policy or whether they have all been constituency-building exercises that produced strong signal but weak action. The honest answer is that it varies. The 2015 letter's research priorities are now standard in safety papers. The 2025 superintelligence letter has shaped the EU AI Act conversation more than the Act's drafters would admit. The 2026 Pro-Human Declaration is too new to evaluate.
What is clear is that the open letter remains a useful tool, but only when it is specific, when it draws a real coalition, and when it offers a direction rather than just a stop.
The format's structural limits
Open letters have a credibility problem that no amount of signatories can fully solve. The format rewards concision, which means complex positions get reduced to slogans. It rewards consensus, which means the most defensible positions are the most general. And it rewards signal-of-concern over signal-of-solution, which means the letters that get the most attention are often the ones least likely to produce action.
The future of the format, if it has one, is probably specialization. The superintelligence prohibition letter worked in part because it was specific to a particular capability threshold. The Pro-Human AI Declaration worked in part because it offered a specific alternative direction. Letters that try to do everything end up doing nothing.
The 2026 letter ecosystem is in a healthier place than the 2023 one, partly because the questions have gotten more specific. The conversation has moved from "should we pause" to "what should the Standards Body look like," and that is a more productive frame. The next round of open letters will be most useful if they continue that pattern of specificity, and most likely to fail if they return to the abstract register that produced the 2023 pause letter's mixed legacy.