The last weekend of May brings a concentrated burst of AI news, led by OpenAI's Rosalind Biodefense initiative — a program that gives vetted developers and government partners restricted access to GPT-Rosalind, its biology-specialized frontier model, for biodefense and pandemic preparedness.

Rosalind Biodefense: Defensive Acceleration

Announced Friday, Rosalind Biodefense operationalizes GPT-Rosalind (launched mid-April) for a narrow set of defensive use cases: epidemiological modeling, biosurveillance, early detection, medical countermeasure development, and outbreak response. Access is sponsored or free for approved organizations.

Initial partners include the U.S. Center for AI Standards and Innovation (CAISI), the U.K. AI Security Institute (UK AISI), Los Alamos National Laboratory, and the Frontier Model Forum. OpenAI's framing is "defensive acceleration" — ensuring that frontier AI advantages defenders more than it advantages adversaries.

The model, named after crystallographer Rosalind Franklin, has been available since April for enterprise and research customers (Amgen, Moderna, and others). The biodefense track is a separate, more restricted channel.

Why It Matters

This is the first time OpenAI has carved out a dedicated biodefense access program for a domain-specific frontier model. It joins Anthropic's Project Glasswing (cybersecurity, with Claude Mythos) and Google DeepMind's AlphaGenome (genomics) in a coordinated pattern: all three frontier labs now operate restricted-access programs for the highest-risk domains — cybersecurity and biology — where the dual-use risk of unrestricted access outweighs open availability.

Codex Pro Promo Expires Today

Meanwhile, the introductory 2x capacity multiplier on Codex Pro ($100/month) ends May 31. From June 1, subscribers' effective monthly capacity halves unless OpenAI announces an extension. This is the third promo expiry in 10 days: Cursor's 2x rate ended May 25; Opus 4.7's GitHub Copilot promo expired April 30.

The pattern — launch with a time-limited multiplier, then normalize pricing within 90 days — is the industry standard in 2026. Teams that built workflows during the promo window at subsidized capacity should model their usage against standard rates starting tomorrow.

Microsoft Build 2026: Windows Becomes the Agent Platform

Microsoft Build kicks off Monday, June 2–3 in San Francisco. Pre-event communications have confirmed the core announcements:

  • Windows Agent Framework — APIs for running autonomous AI agents directly inside Windows. Insider preview in June.
  • WSL 3 — A re-architected Linux subsystem with near-native GPU/NPU access, effectively closing the AI development gap for Windows users.
  • Azure Agent Mesh — A federated control plane for agent execution across on-prem, cloud, and edge. GA targeted for Q4 2026.
  • Windows Agent Store — A marketplace for AI agents with an 85% developer revenue share; Adobe and Zoom are early design partners.

Satya Nadella's keynote will position Windows as the premier platform for the AI agent era — the same strategic playbook Microsoft used to dominate enterprise software in the 1990s.

SoftBank Commits Billions to European AI Infrastructure

SoftBank Group announced a joint venture with Sesterce to build and operate a 1 GW AI data center campus in Bosquel, Hauts-de-France. The initial commitment is €45 billion over five years for 3.1 GW of capacity, with a potential expansion to €75 billion and 5 GW. Facilities are expected online between 2028 and 2031.

The Bottom Line

The dominant narrative today is the shift of frontier AI from private labs into national security infrastructure. OpenAI's Rosalind Biodefense program is the clearest signal yet that biology models are now treated as dual-use technologies requiring government-grade access controls. Combined with Microsoft's agent-platform bet and SoftBank's infrastructure mega-deal, the message from May 2026 is consistent: AI is no longer just a product category — it's becoming a strategic asset managed at the sovereign level.