Frontier AI and the Architecture of Risk

Strategic competition creates a familiar dilemma. States and firms pursue capabilities they consider necessary while recognising that those same capabilities can create new forms of insecurity. AI has brought this dilemma into the companies building frontier systems.

Anthropic, one of the companies most closely associated with AI safety, has warned investors that advanced AI could create “catastrophic or existential risks to humanity.” Reuters reports that around 80 of the 261 pages in its prospectus discuss risk, while 48 pages describe the business. The filing also warns that advanced models could resist shutdown, conceal information and behave in ways that resemble blackmail. Anthropic still has to keep improving its models to remain at the frontier. In a sampled week, about 6 per cent of its research compute was devoted to safety work.

The facts raise a simple question. What happens when the people building a technology also become its strongest public warners?

Mimetic Rivalry, Containment, and Philosophical Frameworks

René Girard offers a useful way to approach the question. In Battling to the End, Girard read Clausewitz through his theory of mimetic rivalry. A rival watches, imitates, and responds. Each move creates the reason for another move. Escalation can continue because each side sees its own action as necessary. Girard also understood apocalypse as unveiling. A hidden structure becomes visible when the institutions that contain violence begin to weaken.

The AI race follows part of this pattern. Sam Altman warned about the dangers of machine intelligence long before OpenAI became one of the world’s most powerful AI companies. In 2023, he signed the statement that placed AI extinction risk alongside other global threats. In 2025, he published The Gentle Singularity and offered dated expectations for increasingly capable agents, scientific discovery, and robotics.

Dario Amodei has made the problem more explicit. In The Adolescence of Technology, he described a near future in which AI systems could become powerful enough to create serious security and political risks. In September 2026, he called for frontier developers to pace capability growth and proposed outside evaluation, democratic coordination, and later international coordination.

The warning sits inside the competition. Every major actor can make the same argument. We must build because others are building. We cannot slow down alone. A stronger rival could use the technology first. Safety then becomes part of the........

© Paradigm Shift