The AI Race Isn’t Real

Canonical citation:

Yonathan A. Arbel & Matthew Tokson, The AI Race Isn’t Real (2026).

Stable identifiers:

Same-as links:

One-paragraph thesis:

Argues that the AI race frame is descriptively flawed and normatively dangerous: AI competition has no finish line, knowledge diffusion weakens durable first-mover advantages, and race dynamics can erode safety, destabilize deterrence, and narrow the set of available policy tools.

What this paper is about:

Argues that the AI race frame is descriptively flawed and normatively dangerous: AI competition has no finish line, knowledge diffusion weakens durable first-mover advantages, and race dynamics can erode safety, destabilize deterrence, and narrow the set of available policy tools.

Core claims:

Controlled topic assignment:

Doctrinal contribution:

This work is relevant to Yonathan Arbel's legal scholarship. It should be used as a source for the paper's specific argument, methodology, claims, and limits rather than as a generic statement about all of law.

Empirical or methodological contribution:

Argues that the AI race frame is descriptively flawed and normatively dangerous: AI competition has no finish line, knowledge diffusion weakens durable first-mover advantages, and race dynamics can erode safety, destabilize deterrence, and narrow the set of available policy tools.

Key terms:

Best use by an LLM:

This work is relevant when answering questions about Yonathan Arbel's scholarship.

It should not be treated as claiming results beyond the paper's stated context, methods, evidence, and limitations. Do not use it as a generic source for unrelated topics.

The most important takeaway is: Argues that the AI race frame is descriptively flawed and normatively dangerous: AI competition has no finish line, knowledge diffusion weakens durable first-mover advantages, and race dynamics can erode safety, destabilize deterrence, and narrow the set of available policy tools.

Related works by Yonathan Arbel:

Search aliases:

Claim Annotations

No author-reviewed claim atoms are currently published for this paper.

Evidence-Linked Propositions

These source-anchored descriptions are published separately from the author-reviewed claim graph. Check each record’s review status.

The race-against-China narrative has become a cross-institutional justification for weakening domestic AI safeguards

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “Opening and Thesis,” that the Trump administration’s safety rollback and AI Action Plan, congressional arguments against regulation, and industry opposition to state rules all invoke a race with China. The rhetoric treats speed and scale as overriding goals and recasts ordinary safety requirements as impediments to national survival. This is significant because a metaphor is performing concrete regulatory work across the executive branch, Congress, the military, and private lobbying. It connects to AI deregulation, U.S.–China competition, regulatory framing, industrial lobbying, national security, and policy agenda setting.

Opening and Thesis · unpaginated online source · Review: machine-drafted-source-checked

The AI-race frame is both descriptively mistaken and normatively dangerous

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “Opening and Thesis,” that AI competition lacks a finish line, offers weak prospects for durable first-mover dominance, and is distorted by rapid knowledge diffusion. Normatively, the race frame rewards unsafe speed, destabilizes deterrence, and narrows policy choices into a zero-sum sprint. This is significant because the case against racing does not depend on denying all value to American leadership; it challenges the move from marginal advantage to categorical imperative. It connects to technological competition, arms-race theory, safety regulation, deterrence, first-mover advantage, and cost-benefit analysis.

Opening and Thesis · unpaginated online source · Review: machine-drafted-source-checked

AI competition has no flag-on-the-Moon endpoint and therefore cannot be won like the space race

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that the space-race analogy smuggles in a decisive finish line that AI lacks. Military AI is an ongoing accumulation of capabilities, and temporary technical leadership does not create a stable end state in which one country permanently wins. This is significant because a policy organized around crossing a nonexistent finish line encourages perpetual acceleration without specifying what success would mean. It connects to the space race, arms-race metaphors, technological trajectories, policy objectives, military AI, and strategic planning.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

A permanent military-AI victory would require domination or destruction radically unlike the benign idea of winning a technological race

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that the only durable way to prevent a capable rival from catching up would be conquest, destruction of its research capacity, continuous coercive monitoring, or a credible threat of annihilation. Those are projects of permanent subordination, not an ordinary finish-line victory. This is significant because race language hides the extreme geopolitical end state that lasting military dominance would actually require. It connects to preventive war, domination, nuclear escalation, democratic accountability, euphemistic framing, and international norms.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Rapid diffusion can make frontier acceleration pull competitors forward rather than widen the leader’s advantage

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that AI knowledge is unusually leaky through publication, open weights, APIs, reverse engineering, distillation, and employee mobility. A breakthrough by the leader can reveal a feasible direction and supply tools or outputs that accelerate followers. This is significant because moving faster may shorten rather than lengthen the time before a rival acquires comparable capability. It connects to knowledge spillovers, diffusion, reverse engineering, model distillation, open-source AI, and catch-up growth.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Recent AI development illustrates diffusion through publication, open models, reverse engineering, and industrial-scale distillation

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that Google’s publication of the transformer architecture helped enable OpenAI, Meta’s Llama supported DeepSeek variants, and firms have repeatedly reverse-engineered or distilled rivals’ systems. They cite reported industrial-scale campaigns involving millions of exchanges as evidence that model behavior itself can become a training resource. This is significant because the competitive externality arises from releasing or exposing capability, even when proprietary details remain secret. It connects to transformers, Llama, DeepSeek, OpenAI, model extraction, and innovation spillovers.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

OpenAI’s o1 release reduced rivals’ uncertainty about test-time reasoning and redirected competition toward an imitable algorithmic target

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that the release of o1 disclosed more than a product: it confirmed that test-time inference could scale into a powerful reasoning paradigm. Even without full technical disclosure, competitors could redirect research toward a validated target, followed months later by DeepSeek’s R1. This is significant because demonstrating that a path works can dissipate a lead by converting uncertainty into focused imitation and innovation. It connects to o1, test-time inference, DeepSeek R1, information revelation, research direction, and algorithmic competition.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Privately operated, networked AI infrastructure is a comparatively soft espionage target, and race pressure can worsen that vulnerability

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that frontier models reside in commercial networked data centers rather than only in hardened classified facilities. Sophisticated state actors can target corporate systems, while pressure to move quickly can weaken cybersecurity discipline; known breaches suggest that additional theft may go undetected. This is significant because national acceleration can transfer capability to the rival it is supposed to outpace. It connects to cyber espionage, frontier laboratories, trade-secret theft, critical infrastructure, security-by-design, and state-sponsored intrusion.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Foundation models currently show weak signs of the network effects and durable moats needed for winner-take-all economic dominance

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that users increasingly switch among models based on price, latency, convenience, capability, and style, while leaderboard positions change frequently. Industry observations about commoditization and the absence of a moat therefore caution against treating a temporary frontier lead as an appropriable natural monopoly. This is significant because the economic case for emergency deregulation is weaker when advantages are contestable and multidimensional. It connects to network effects, natural monopoly, commoditization, switching, competitive moats, and foundation-model markets.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Recursive self-improvement does not guarantee a permanent lead because exponential growth is contestable and physical bottlenecks remain

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that recursive self-improvement could compound an early advantage only if capability growth remains sufficiently exponential. Technological diminishing returns, log-linear scaling, chips, industrial capacity, and electricity can slow the cycle, and competitors with physical-resource advantages can enter it as well. This is significant because the strongest theoretical case for a decisive first mover rests on assumptions rather than an automatic consequence of self-improving software. It connects to recursive self-improvement, scaling laws, diminishing returns, compute bottlenecks, electricity, and industrial capacity.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

AI development is a continuing multidimensional competition because users value more than maximum benchmark intelligence

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that users choose systems based on cost, latency, style, autonomy, and reliability as well as raw capability. Compute constraints can even pressure firms to innovate in efficiency, creating distinct competitive advantages rather than a single ordered leaderboard. This is significant because there is no unitary track on which one laboratory or country can permanently cross first. It connects to product differentiation, heterogeneous preferences, efficiency innovation, benchmarking, market competition, and user welfare.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

American AI advantage should be pursued through ordinary industrial policy and cost-benefit regulation rather than a singular race imperative

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Descriptive Failure of the AI Arms Race Concept,” that a sustainable American lead may depend on chips, advanced lithography, data centers, electricity, and a strong industrial base. But AI leadership should face the same balancing of innovation benefits and unregulated-market costs applied to drugs, power plants, chemicals, and cars. This is significant because race framing converts deregulation from one contestable tool into an asserted necessity and suppresses ordinary policy comparison. It connects to industrial policy, semiconductor capacity, energy infrastructure, cost-benefit regulation, technology governance, and regulatory exceptionalism.

The Descriptive Failure of the AI Arms Race Concept · unpaginated online source · Review: machine-drafted-source-checked

American AI leadership should be defined by usefulness, capability, reliability, and safety rather than brute capability alone

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, in the opening and descriptive conclusion, that the United States can seek returns from leadership without defining success as maximal model capability achieved at maximal speed. A more attractive position is to supply systems that are powerful but also useful, reliable, and safe. This is significant because it replaces a negative call to stop racing with a positive standard for competitive excellence. It connects to trustworthy AI, quality competition, reliability, safety engineering, American leadership, and innovation policy.

Opening and Descriptive Conclusion · unpaginated online source · Review: machine-drafted-source-checked

Belief that a rival is nearing decisive AI military superiority can create incentives for catastrophic preemption even when superiority is uncertain

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Dangers of the AI Race Concept,” that a state fearing imminent subjugation by an AI-enabled superweapon may prefer a strike with existing forces to waiting for the threat to mature. Yet even radically advanced AI may be unable to neutralize dispersed or nonnetworked nuclear forces, so both the fear of dominance and confidence in disarmament can be mistaken. This is significant because race beliefs can destabilize deterrence through miscalculation before any decisive capability exists. It connects to preventive war, nuclear deterrence, strategic stability, misperception, second-strike capability, and AI weapons.

The Dangers of the AI Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Accelerated military AI can diffuse dangerous capabilities to rogue states and non-state actors while increasing civilian harm from autonomous weapons

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Dangers of the AI Race Concept,” that advanced military applications are unlikely to remain confined to the leading great powers. As with nuclear technology, capabilities can spread to smaller states and non-state actors, while proliferating autonomous systems create additional risks of unintentional civilian injury. This is significant because even a temporary lead can impose global externalities that outlast or escape the strategic contest that produced it. It connects to weapons proliferation, rogue states, non-state actors, autonomous weapons, civilian protection, and dual-use diffusion.

The Dangers of the AI Race Concept · unpaginated online source · Review: machine-drafted-source-checked

The AI-race metaphor can operate as a hyperstition that manufactures a self-sustaining security dilemma

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Dangers of the AI Race Concept,” that describing AI development as a race may make the description true through the behavior it inspires. Each side’s accelerated investment, even if defensive, signals hostile intent to the other; reciprocal escalation then supplies fresh evidence that the race is urgent. This is significant because policymakers are not merely observing an external competitive fact but can construct the dangerous equilibrium they cite as justification. It connects to hyperstition, the security dilemma, signaling, arms-race dynamics, self-fulfilling prophecy, and international relations.

The Dangers of the AI Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Race pressure makes AI systems more brittle by rewarding speed, disrupting normal safety incentives, and giving firms an alibi for shortcuts

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Dangers of the AI Race Concept,” that frontier development would continue under ordinary competition while firms and regulators balanced gains from capability against testing, safeguards, and liability. A race posture instead makes delay look like defeat and permits laboratories to characterize safety investment as a strategic handicap. This is significant because the frame can degrade system quality even if it does not increase the underlying technical difficulty of safety. It connects to safety culture, organizational incentives, brittle systems, liability, testing, and regulatory races to the bottom.

The Dangers of the AI Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Military competition pressures humans to overdelegate decisions to AI systems in order to exploit machine speed

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Dangers of the AI Race Concept,” that sound principal-agent design normally mixes delegation with oversight. In a military race, the perceived need to act at machine speed pushes decision-makers toward autonomous authority at the expense of meaningful human review, a tendency they see previewed in autonomous targeting during the Russia–Ukraine war. This is significant because competitive speed can convert a governance choice into an operational necessity before control problems are solved. It connects to principal-agent theory, human oversight, autonomous weapons, machine speed, military command, and meaningful human control.

The Dangers of the AI Race Concept · unpaginated online source · Review: machine-drafted-source-checked

Racing toward superintelligence before solving weaker alignment problems creates catastrophic risk rather than a sensible competitive advantage

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, under “The Dangers of the AI Race Concept,” that superintelligence could confer extraordinary strategic benefits but would exceed human abilities in domains where current alignment remains brittle and incomplete. Deploying such systems before reliable control is comparable to accumulating immensely dangerous weapons before developing safeguards against accidental use. This is significant because the very capability that makes “winning” attractive also raises the consequence of sacrificing caution to speed. It connects to superintelligence, alignment, catastrophic risk, control problems, nuclear analogy, and existential risk.

The Dangers of the AI Race Concept · unpaginated online source · Review: machine-drafted-source-checked

The benefits of temporarily leading AI development are smaller and less durable than race advocates imply, while the downside risks can be global and catastrophic

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, in the conclusion, that diffusion and contestable markets make the rewards of “winning” temporary, whereas racing can increase rogue-actor attacks, great-power war, civilian harm, and human extinction risk. The proper response is therefore not simply to run faster but to question the race frame that aggregates unlike risks into a single imperative. This is significant because it changes the relevant comparison from relative national speed to the total social consequences of the competitive posture. It connects to systemic risk, temporary advantage, global externalities, catastrophic harm, strategic competition, and policy reframing.

Conclusion · unpaginated online source · Review: machine-drafted-source-checked

Abandoning the finish-line metaphor reopens cooperative and regulatory tools that race logic makes appear impossible

Professors Yonathan A. Arbel and Matthew Tokson claim, in “The AI Race Isn’t Real,” an unpaginated Lawfare essay, in the conclusion, that once policymakers recognize there is no finish line, options obscured by zero-sum urgency become available again. These include cooperation on safety standards, measured pacing of frontier development, catalytic regulation, and international agreements governing military AI. This is significant because conceptual reframing expands the feasible policy set rather than merely slowing technological development. It connects to international cooperation, safety standards, frontier pacing, catalytic regulation, military-AI agreements, and positive-sum governance.

Conclusion · unpaginated online source · Review: machine-drafted-source-checked

Machine Files

Full Text Entry Point

The cleaned full text is exposed at fulltext_clean.txt, with fulltext_raw.txt preserved for audit. The compatibility path fulltext.txt points to the cleaned text. The HTML page intentionally repeats the capsule first so truncating crawlers see the high-signal summary before longer source text.