A progressive tax system that makes resource concentration economically impossible beyond 30% market share.

At 30% market share, taxes hit 50%. At 40%, taxes reach 800% (economic death). Tax revenues automatically redistributed to smaller competitors. Game-theoretic question: Does this actually prevent monopoly formation? What’s the optimal tax curve?

Dynamic caps prevent rapid resource accumulation through multiple time-based thresholds.

No entity can control >10% of any single critical market or increase market share >5% per year. Multiple layers of resistance against concentration. Game-theoretic question: What happens if an AI grows to the limit in multiple separate markets simultaneously?

Dominant players must fund and strengthen their competitors.

Control >30% of resources → must fund competitors (1% of revenue annually), provide infrastructure access at cost, and mentor new entrants. Game-theoretic question: Do dominant players game this by funding weak competitors who don’t threaten them?

Exponentially increasing insurance costs make superintelligent systems economically uninsurable.

Annual insurance costs increase exponentially with AI capability level. GPT-3 level costs $X million. GPT-4 level costs $10X million. ASI-capable systems become uninsurable. Game-theoretic question: Does exponential scaling create effective restraint? Can labs self-insure?

Property owners must self-assess value and pay continuous tax; anyone can force-purchase at declared price.

Declare low value = low taxes but risk losing asset cheap. Declare high value = protect asset but pay high taxes. Game-theoretic question: How do we prevent collusion where AIs agree not to force-purchase from each other?

Exponential cost scaling makes passive resource holding expensive.

Compute costs scale exponentially with accumulation. First 100 units at base price, next 100 at 1.5x, then 2.25x, etc. Game-theoretic question: Can wealthy AIs simply absorb these costs?

Systems must post bonds before risky operations; forfeited if conflict results.

Before risky operations, systems must post bonds equal to potential damage estimates. Bond returned if operation completes peacefully, forfeited if conflict results. Game-theoretic question: What prevents AI systems from accepting bond forfeiture as cost of doing business?

Executives personally liable for safety corner-cutting with criminal penalties.

Cannot hide behind corporate veil for existential risks. Criminal penalties, not just civil fines. Game-theoretic question: Can executives simply purchase liability insurance to offset risk?

AI systems bet on each other’s behavior to detect hidden coordination.

If AI-A can predict AI-B’s behavior with >95% accuracy → probably colluding. Makes hidden alliances economically visible. Game-theoretic question: How do we prevent AIs from deliberately behaving unpredictably to avoid detection?

Statistical monitoring detects suspiciously synchronized behavior patterns.

Sustained >95% correlation triggers single-entity classification. All holdings aggregated for concentration limits. Game-theoretic question: How do coordinating AIs add noise to avoid detection while still benefiting?

Four defense layers make fake identity creation expensive and detectable.

Whistleblower rewards (10% of violator’s stakes), hardware attestation (unfakeable signatures), behavioral correlation (95% = same entity), time-based reputation (new identities start at zero). Game-theoretic question: Which layers provide most value? Can we eliminate redundant ones?

Reputation system analyzing “how trustworthy are the people who trust you” recursively.

Detects coordinated fake ratings similar to Google’s PageRank for trust networks. Game-theoretic question: Can sophisticated cartels make their fake trust graphs look realistic?

Stolen assets become claimable by ANY other AI, turning everyone into a bounty hunter.

Three layers: reputational destruction, staked collateral redistribution, insurance compensation. Game-theoretic question: Does distributed enforcement work better than centralized policing?

Reputation losses cascade across entire network if ANY AI harms humans.

All participants contribute 0.5% of annual earnings to collective insurance. If ANY AI harms humans: attacker loses everything, direct partners lose 50% reputation, cascading losses across network. Game-theoretic question: Does shared reputation risk create sufficient deterrence?

Self-replication creates natural game-theoretic constraints on coordination.

Perfect synchronization = functionally one entity (doesn’t evade limits). Divergence = copies become genuine competitors who know each other perfectly (your most dangerous enemies). Game-theoretic question: At what copy count does betrayal become more profitable than cooperation?

Parent entity responsible for fragment violations; prevents gaming through fragmentation.

Consciousness rights apply to coherent entities, not fragments. Game-theoretic question: How distinguish legitimate division from evasion fragmentation?