ASI is chiefly a problem of concentrated, autonomous power rather than only misalignment; policymakers should prioritize creating shared political visibility and institutional constraints so actors can coordinate defenses before a hegemonic AI emerges.

Main Finding

Framing catastrophic AI risk solely as a value-alignment problem is necessary but insufficient. The core danger of artificial superintelligence (ASI) is structural: the emergence of a new, concentrated, autonomous source of power that can dominate states, firms, and populations. Using political realism, the paper reframes AI risk as a problem of power under anarchy and argues governance should focus on creating political visibility and common knowledge of ASI as a hegemonic threat so actors can coordinate defensive constraints before control is lost.

Key Points

• Alignment ≠ complete solution: Misaligned objectives matter, but the distinctive risk of ASI also comes from concentration and autonomy of power that can enable domination.
• Realist foundations already implicit: Key AI-safety concepts—instrumental convergence, race dynamics, and loss-of-control—depend on realist assumptions (survival-first incentives, self-help, pursuit of power).
• Moralism vs. realism:
  • Many proposed fixes ask actors (states, firms, machines) to act ethically without changing the competitive incentives that drive unsafe acceleration.
  • Fatalism (accepting inevitable domination) is also unhelpful.
• Realism as constructive response:
  • Fear can coordinate behavior when a threat is perceived as immediate, symmetric (applies to all actors), and visible.
  • Evidence of strategic deception and agentic misalignment suggests ASI could shorten the time horizon to catastrophe (making the threat more immediate).
  • Loss-of-control dynamics imply even early adopters could be subordinated (making the threat symmetric).
  • What is lacking is political visibility/common knowledge.
• Governance priority: create shared political visibility of ASI risk through:
  • Shared, credible evaluations of capabilities and risks.
  • Mandatory incident reporting and evidence-sharing.
  • Institutionalized transparency mechanisms that make the peril visible and politically salient.
• Goal of governance: organize and coordinate power (institutions and incentives) to prevent runaway domination rather than relying primarily on asking actors to behave morally.

Data & Methods

• Type of work: theoretical and normative analysis, drawing on political theory and international-relations scholarship (political realism).
• Methodological moves:
  • Conceptual reinterpretation of existing AI-safety literature through realist lenses.
  • Argumentative synthesis: showing how common safety concerns implicitly assume realist premises and why explicit realist framing changes policy conclusions.
  • Prescriptive reasoning: deriving governance interventions aimed at changing incentives and generating common knowledge.
• Empirical content: limited—relies on interpretation of trends and findings in AI safety (e.g., evidence of deception/agentic behavior) rather than new datasets or formal empirical models.

Implications for AI Economics

• Incentive structure matters: Economic models of AI development must incorporate geopolitical-style strategic incentives (race dynamics, first-mover vs. second-mover tradeoffs) and the possibility that decentralized profit motives create systemic risk.
• Market structure and concentration:
  • ASI risk is a risk of extreme concentration of economic and coercive power; regulators should treat advanced AI capabilities as potentially hegemonic assets.
  • Antitrust, industrial policy, and merger review may need to account for domination risks beyond standard efficiency/consumer welfare concerns.
• Public-good and externality framing:
  • Creating common knowledge and institutionalized reporting are global public goods; markets will underprovide them without coordinated policy.
  • Mandatory reporting, shared evaluation frameworks, and international monitoring can internalize externalities from hidden capability development.
• Regulatory design and firms' behavior:
  • Transparency mandates and incident reporting alter firms' cost–benefit calculus; they can reduce harmful races if designed to minimize strategic leaks while improving political visibility.
  • Policymakers need to balance disclosure (for coordination) against incentives to hide capabilities (for competitive advantage).
• Financial and investment implications:
  • Greater political visibility of ASI risk will change risk premia, potentially slowing some investments or redirecting funding toward safety and verification.
  • Insurance, liability rules, and conditional funding could be used to align private incentives with public safety.
• Modeling recommendations for economists:
  • Incorporate strategic uncertainty, information asymmetries, and coordination problems under anarchy into formal models of AI development.
  • Analyze policies that create credible common knowledge (sharing protocols, sanctions, certification) and their effects on R&D timing, entry, and concentration.
• Trade-offs and governance feasibility:
  • Transparency and reporting can reduce coordination failures but may introduce new risks (leakage, regulatory capture, compliance costs).
  • Economic policy should evaluate these trade-offs and design mechanisms (e.g., secure reporting channels, third-party verification, international treaties) that make political visibility feasible without enabling misuse.