OpenAI and Anthropic publish second-round results from their joint cross-lab safety evaluation — establishes cross-lab evaluation as a permanent fixture of frontier-model alignment infrastructure
OpenAI and Anthropic publish their second round of joint cross-lab safety evaluations on each other's publicly released models. The cadence (now twice within 9 months) establishes cross-lab evaluation as a permanent alignment-infrastructure fixture — and validates the pattern METR is operationalizing across the larger four-lab perimeter (yesterday-PM news).
The substantive piece is the cadence-establishment signal. The first OpenAI-Anthropic joint evaluation in late 2025 was treated as an experiment; the second round in mid-2026 establishes the cadence as a permanent fixture. The two-lab structure complements the METR cross-lab pilot (Anthropic, Google, Meta, OpenAI internal-developer agents) — together they produce a two-tier cross-lab evaluation infrastructure where direct-bilateral (OpenAI-Anthropic) covers the highest-stakes evaluations and METR-mediated cross-lab covers the broader four-lab perimeter.
The structural read against OpenAI's Deployment Simulation announcement is that the alignment-infrastructure tier is solidifying around three primitives: replay-evaluation (production-data-driven pre-launch testing), cross-lab evaluation (direct-bilateral and METR-mediated), and internal-agent evaluation (METR pilot, Anthropic Risk Report). The H2 2026 alignment posture for the field is the most operationally-mature it has ever been — capability-side pace is being matched by alignment-infrastructure operationalization for the first time.
OpenAI — Findings from a pilot Anthropic-OpenAI alignment evaluation exercise → · Anthropic — Research → · Anthropic — Core views on AI safety →