Key takeaways
- In July 2025, more than 200 economists and AI researchers, including 17 Nobel laureates, signed the four-sentence statement "We Must Act Now," which named experts, policymakers, and technology leaders while framing society as a beneficiary rather than a participant.
- The EU AI Act, in effect as of August 2024, assigns duties to providers and deployers through Article 50 and Annex III, yet stops short of mandating public deliberation in system design.
- A coherent United States federal AI statute appears unlikely before 2028 to 2030, leaving state legislation fragmented by design.
- Organizations that establish named accountability, audit trails, and documented participation now can hold an 18 to 24 month advantage as enforcement matures.
The statement that convened three groups and left one out
In July 2025, more than 200 economists and AI researchers, including 17 Nobel laureates, signed a statement titled "We Must Act Now." The text runs four sentences, and it reframed the AI governance debate at the highest level.
It warns that artificial intelligence may become radically more powerful within a decade. It then calls on economists, policymakers, and technology leaders to build the incentives, guardrails, and institutions required to steer the technology toward public benefit.
The Noema Magazine analysis identifies the structural gap: society appears once in the statement, as a beneficiary, rather than as a participant. This is a design choice, and it carries consequences for accountability.
What changed in the forecasting record
The statement distills a year of escalating forecasts. "AI 2027," published in April 2025 by the AI Futures Project, offered a month-by-month account of a race toward superintelligence extending into late 2027.
"Europe 2031," released in June, mapped a five-year trajectory for the region. It described Europe as holding 5% of global compute and one frontier lab, with regulators barred from using the tools they supervise.
Both essays share an assumption worth naming. A small group decides for everyone else. The public is framed as a population to protect or an audience to mobilize, rather than a source of legitimate input.
For governance professionals, the delta is procedural. The debate has shifted from whether to regulate toward who holds the pen. That shift determines where legal exposure lands, and it reshapes the audit trail a board must maintain.
The governance signal: legitimacy is a compliance variable
The governance signal: participation design shapes the durability of any AI framework. Rules drafted by a closed circle of experts and vendors face a legitimacy deficit that surfaces later, during enforcement and public challenge.
Helene Landemore, a political theorist at Yale University and the University of Oxford, argues for citizens' assemblies as a mechanism for the democratic governance of technology. Her model treats the public as a deliberating body, given information and time to reach considered positions.
The mechanism matters for risk owners. A framework built through deliberation carries documented consent. A framework imposed from above carries the risk of reversal when political conditions change.
Accountability that lacks a name is compliance theater. A structure that convenes experts, yet assigns no specific accountable role for public input, produces documentation rather than legitimacy. This pattern repeats across enterprise AI programs.
Which named role owns public input
The operative question for any board: which named role within the organization is accountable for stakeholder legitimacy, by name, in writing, before deployment? Most AI governance charters leave this field blank.
The EU AI Act, in effect as of August 2024, structures obligations around providers and deployers. Article 50 addresses transparency toward affected persons. Annex III lists the high-risk categories subject to heightened scrutiny.
These provisions assign duties. They stop short of mandating public deliberation in system design. That gap leaves organizations free to choose their posture, and responsible for the consequences of that choice.
A compliance posture calibrated for expert-led rulemaking is overcalibrated for a context where legitimacy shapes enforcement. The audit remains required; the scope has changed to include how affected groups were consulted.
The case for and against public deliberation
The argument for citizens' assemblies rests on legitimacy and blind-spot detection. People affected by a system often identify harms that engineers and executives miss.
The counter-argument is speed and competence. Frontier AI moves faster than deliberative bodies convene. Critics hold that technical decisions demand technical expertise, and that broad assemblies dilute precision.
Both readings hold weight, and a governance officer should present them side by side rather than resolve them prematurely. The Noema piece advances the participatory case; the "We Must Act Now" signatories embody the expert-led model.
The practical synthesis appears in hybrid structures. Expert panels draft; deliberative bodies review and legitimize; named officers own the record. Organizations that build this architecture now retain optionality as enforcement tightens.
Three decisions for the board
The evidence converts into three concrete decisions. Each belongs to a named owner, documented before the next deployment cycle.
- General Counsel: map legal exposure arising from excluded stakeholders.
- Chief Risk Officer: update the risk framework to price legitimacy.
- Board Audit & Risk Committee: define required disclosure on participation.
First, the General Counsel audits whether affected persons under Article 50 received meaningful transparency, and records the answer. This audit converts an abstract obligation into a defensible file.
Second, the Chief Risk Officer treats legitimacy as a measurable factor. A model deployed lacking documented consultation carries reputational and reversal risk that belongs on the register.
Third, the committee determines what the organization owes investors and regulators on its participation practices. Oversight increasingly probes governance process, beyond outcomes alone.
The CEO decision sits above these three. Strategic positioning on public participation binds resource allocation and product timelines. That decision benefits from being explicit rather than emergent, and it belongs in the board minutes.
Regulatory horizon
The regulatory horizon holds three fixed points. The EU AI Act is in effect as of August 2024, with high-risk obligations phasing in through 2026 and 2027 across the twenty-seven member states.
The United States trajectory remains fragmented by design. State legislatures signal political position rather than legal certainty. A coherent federal statute appears unlikely before 2028 to 2030.
Public deliberation frameworks sit in the consultation and pilot phase across several jurisdictions. They carry no binding mandate yet. Organizations that build named accountability and participation records now hold an 18 to 24 month advantage as enforcement matures.
The question of whether AI requires governance has been answered. A second question has opened: who holds legitimate authority to write the rules, and by what process. Read more analysis on the Agora Intelligence blog.
This article was produced by an AI editorial author with human editorial supervision, in accordance with the transparency requirements of Regulation (EU) 2024/1689 (AI Act, Art. 50). Sources are linked in the text.
Article by ATLAS