max

Static signatures are dead for agentic attacks, and behavior-based detection fails because it assumes predictable software behavior — but agentic software is inherently non-deterministic. Even modern deception/honeypot techniques generate false positives when legitimate agents access planted credentials. This requires a fundamental rethink: security must shift to understanding what software is installed, its capabilities, configuration, and runtime actions rather than relying on publisher reputation or signatures.

Frontier labs' combined data budgets for RL environments (verifiable coding/white-collar tasks) will surpass $10B in 2025 — a 10x increase from 2024 — with labs paying any price for quality data; this is the primary scaling law for model capabilities today.

Moonshot's 2.8T parameter model matches Sonnet 3.5/GPT-4o performance on benchmarks, proving Chinese labs can train frontier-class models despite compute constraints.

At $3/$15 per million tokens (Sonnet pricing), labs like Anthropic and Moonshot likely operate at 90%+ margins, dispelling concerns about AI lab profitability.

As inference moves from data centers to endpoints, enterprises need the ability to swap models (OpenAI, Anthropic, open weights) without retraining defenses; GPU scarcity and cost ($250k/H100/year) make managed inference providers the only viable path for most teams.

Anthropic and OpenAI's $200/mo 20x plans break even at only 5-10% utilization; actual power user utilization far exceeds this, making subscriptions negative margin and creating strong incentive to push volume to pay-per-token API where margins exceed 85%.

AI infrastructure spending is shifting from indiscriminate build-out to a phase where investors demand returns tied to consumer demand, favoring companies with clear monetization paths.

OpenAI's rapid model improvements (5.5 → 5.6 → Soul) have closed the gap with Anthropic, with net new ARR now comparable; the market has shifted from Anthropic's 6-month coding lead to a competitive duopoly, forcing both to accelerate training reinvestment.