STRATEGYMonths to result87% confidence

AI Governance Deficit — The Three Missing Institutions

AI is developing faster than governance: a pre-2008 financial system with systemic contagion risk

Problem it solves

Why AI sector political risk is underpriced and what institutional responses are most likely

Best for

Understanding regulatory risk for AI companies and the political trajectory of AI sector sentiment over 2-5 year horizon

Not ideal for

Timing specific regulatory events or trading on near-term policy catalysts

Overview

Why this framework exists

Bremmer's AI Governance Deficit framework identifies three critical institutions that do not yet exist but are urgently needed to prevent a systemic governance failure in AI. The analogy is the pre-2008 financial system: everyone benefited from the lack of oversight until the contagion became unstoppable. The three missing institutions are: (1) US-China AI arms control, analogous to post-Cuban Missile Crisis nuclear treaties; (2) an AI Stability Board modeled on the Financial Stability Board created post-2008; and (3) a Universal AI Access Fund to prevent a permanent human bifurcation between AI-empowered and AI-excluded populations.

Bremmer treats the governance deficit as a genuine systemic risk, not a theoretical concern. The trigger for the AI Stability Board argument was an Anthropic unreleased model capable of identifying exploitable software vulnerabilities across all banking and critical infrastructure systems — a capability that prompted emergency meetings between Jerome Powell, Scott Bessent, and major bank CEOs. Bremmer cites this as the moment the financial system recognized the AI governance gap was no longer abstract.

The political consequence is already manifesting: AI is less popular than ICE (immigration enforcement) with the US public, a US senator with pro-business centrist credentials told Bremmer he has never seen constituents as upset about any issue as data centers, and AI CEO targeting has begun. The pattern Bremmer identifies is a 2-3 year window before the white-collar unemployment wave from AI creates a 2028 political alignment that drives serious regulatory and tax pressure on AI companies in Western markets.

Core principles

5 total
  1. Technologies that develop faster than governance create systemic risk even when individually beneficial, because the contagion mechanism is invisible until it triggers.
  2. US-China competition in AI without arms control is structurally analogous to 1955-1962 nuclear competition — the period of maximum danger before both sides accept the need for constraints.
  3. AI unemployment hitting white-collar educated workers creates a politically distinct and more electorally powerful backlash than manufacturing displacement.
  4. Without universal AI access, the gap between AI-empowered and AI-excluded populations becomes a permanent human bifurcation that erodes social cohesion.
  5. Emergency institutional responses (like the FSB post-2008) are possible but require a near-catastrophic triggering event first.

Steps

5 steps
  1. Diagnose the governance lag
    Identify the gap between AI capability advancement speed and institutional response speed. Bremmer's diagnostic: we are in the pre-Cuban Missile Crisis phase for AI — the most dangerous period of maximum capability growth with minimum governance constraint. The institutions needed do not exist and are not in active negotiation.
    Pro tipThe absence of arms control discussions, not the capability level alone, is the risk indicator. Nuclear parity existed for years before governance caught up; the gap period is when accidents happen.
  2. Monitor the triggering event pipeline
    Governance institutions are typically built in response to near-catastrophic events rather than in anticipation. Track events that could serve as the governance trigger: an AI-enabled cyberattack on critical infrastructure, a major white-collar employment shock that creates political tipping point, or a publicized AI safety incident at scale.
    Pro tipThe Anthropic unreleased model disclosure is a preview trigger. It prompted emergency institutional response (Fed/Treasury/JPMorgan) but stopped short of a systemic event. Track subsequent disclosures of similar capability.
  3. Track the political sentiment leading indicators
    Bremmer identifies three leading indicators for the AI political backlash wave: public approval polls (AI is already less popular than ICE), constituent anger at elected officials (the senator data point), and physical targeting of AI executives. When all three indicators are active simultaneously, the regulatory response is 1-2 electoral cycles away.
    WarningThe political backlash constituency Bremmer identifies — educated, urban, white-collar, advanced degree holders worried about jobs — is different from the 2016 populist wave. It is the democratic party's core base. When this group turns against AI, the political path for regulation changes completely.
  4. Assess each missing institution's probability and form
    For each of the three missing institutions (arms control, stability board, access fund), estimate separately: probability of being created, likely trigger event, form it would take, and timeline. The FSB analog is the most institutionally mature precedent and most likely to form first following a near-systemic AI security event.
    Pro tipThe FSB was created within 18 months of the 2008 crisis. An AI Stability Board following an analogous triggering event could form on similar timelines — fast by institutional standards.
  5. Translate governance risk into sector impact by actor type
    Different AI actors face asymmetric regulatory risk. Centralized large-model providers face the greatest systemic-risk scrutiny. Decentralized infrastructure providers face different risks under arms control frameworks. Infrastructure providers (data centers, chips) face local political hostility from the constituent-anger channel. Map the impact differently across the stack.
    WarningAn AI Stability Board modeled on the FSB would likely create regulatory moats for established centralized providers over decentralized challengers — the same pattern seen in financial regulation post-2008.

Checklist

Saved in your browser

Examples

2 cases
Anthropic unreleased model — the near-trigger event

An Anthropic model not yet released was found capable of identifying exploitable software vulnerabilities across banking systems and critical infrastructure. Jerome Powell and Scott Bessent called an emergency meeting of major bank CEOs. Jamie Dimon described it as a five-alarm fire. The institutions took it seriously enough to discuss internal deployment as a defensive measure.

OutcomeNo governance institution was created, but the episode demonstrated that the systemic risk is real, recognized at the highest levels, and that the financial sector is treating AI capabilities as a critical infrastructure threat requiring immediate response.
The 2016 to 2028 political wave pattern

The 2016 political wave targeted free trade and immigration as culprits for manufacturing worker displacement. Bremmer identifies the 2028 equivalent: educated white-collar workers — women with advanced degrees in urban and suburban areas — facing AI-driven job displacement and blaming both AI companies and the politicians who enabled them.

OutcomeBremmer calls this as the most likely political scenario in the US for 2028, representing a left-populist wave analogous in structure to the 2016 right-populist wave but targeting a different sector and constituency. It will produce a political environment hostile to AI companies comparable to the 2016-2020 hostile environment for trade agreements.

Common mistakes

4 traps
Treating AI public sentiment as uniformly negative tech skepticism
The political hostility Bremmer identifies is structurally different from past tech-skepticism cycles. It targets educated white-collar workers — the group most politically active and most able to translate grievance into electoral outcomes. Category rejection of AI based on the prior tech-skepticism pattern underestimates the political force of this specific constituency.
Dismissing the Anthropic model disclosure as marketing or hyperbole
Bremmer explicitly states he takes the systemic cybersecurity risk seriously because of the nature of the response: the Fed Chair and Treasury Secretary called an emergency meeting of major bank CEOs. This is not a standard policy briefing; it is a crisis-response activation. The institutional response is the signal, not just the capability claim.
Assuming governance lag is permanent
The FSB post-2008 showed that governance institutions can form rapidly in the wake of systemic events. The risk is not that governance never catches up — it is that the catch-up response after a triggering event will be blunt, fast, and potentially poorly designed. Pricing in governance lag as a permanent feature misses the whiplash risk.
Conflating US-China AI competition with arms control impossibility
Bremmer explicitly states he has back-channel conversations with both sides. The argument is not that arms control is impossible, but that we are in the pre-Cuban Missile Crisis phase where the political conditions for negotiation do not yet exist. The Cuban Missile Crisis created those conditions within days; an AI equivalent event could do the same.

Origin story

How this framework came to be

Bremmer developed this framework through direct access to policymakers and AI company leadership. He explicitly distinguishes himself from commentators who analyze AI from the outside — he has spoken directly with Anthropic, had conversations about the unreleased model disclosure with individuals close to the Fed and Treasury emergency response, and advises governments on AI policy frameworks globally.

The framework draws on his 30 years of risk assessment methodology applied to a new domain. The core method is to identify when a high-consequence technology or institution has outrun its governance architecture and assess the probability and form of the corrective response. He applied this method to financial derivatives pre-2008 (predicting systemic failure) and applies it to AI now with the same structural diagnosis.

Source

Traced to primary
Source · PODCAST
The Global Politics Expert: The Real Global Danger is What Comes Next!
Ian Bremmer · 2025
Open source →

Related frameworks

Browse all Strategy →