dario amodei

Dario Amodei predicts training runs will reach $10B in 2025-2027, and with continued algorithmic and chip improvements, models could exceed most humans at most tasks — a concrete timeline for transformative AI capability.

Anthropic's revenue is growing 3x quarterly (80x annualized) far exceeding its 10x/year compute plan, creating temporary server strain; however, compute market is liquid with Google/Amazon deals and new capacity available within months.

Anthropic's Mythos model autonomously traverses cyber kill chain, finding thousands of critical vulnerabilities; gradual release to defenders aims to patch finite attack surface before model capabilities proliferate to open source.

Anthropic deliberately withheld its most powerful cyber model (Mythos) to develop stronger defenses, costing hundreds of millions in revenue; red lines against mass surveillance and fully autonomous weapons are enforced even with defense partners.

Amodei reveals Anthropic can identify discrete features representing complex concepts (bias, hesitation, metaphors) inside models, enabling detection of memorization vs reasoning and compliance with EU right-to-explanation rules — a practical breakthrough for enterprise trust.

Enterprise model (vs ad-driven consumer) aligns with safety values — long-term trust relationships, curing diseases, energy, education; coding/enterprise focus yields sticky revenue and synergistic safety culture.

Anthropic's RSP — measuring models for misuse and autonomy risks at each compute milestone — has been replicated across the frontier lab ecosystem, creating a de facto self-regulatory layer that may shape future legislation.

AI first makes workers 10x more productive (automating 90% of tasks), then approaches full automation; Anthropic pushes enterprise customers toward 'do more with same resources' not cost-cutting, but transition speed creates dislocation risk.