# Daily Agent Platform Research Briefing — June 15, 2026

Welcome to the agent platform research briefing for Sunday, June 15th. Three stories today.

**Anthropic export controls: White House says lift ASAP after fix** — The week's biggest saga is reaching a potential resolution. After the US Commerce Department barred foreign access to Anthropic's Fable 5 and Mythos 5 models on June 12th, David Sacks — co-chair of the President's Council of Advisors on Science and Technology — posted over the weekend that the administration wants the export controls lifted "as soon as possible" once Anthropic patches the jailbreak vulnerability. The key new detail: Amazon CEO Andy Jassy was among tech leaders who flagged the vulnerability to Trump administration officials, and Amazon researchers originally discovered the jailbreak. Anthropic pushed back publicly, calling the recall standard unreasonable and warning it would "essentially halt all new model deployments across the industry" if applied broadly. The administration called Anthropic's defense "at odds with their branding as a safety-first company." Anthropic has sent senior staff to Washington to negotiate the fix and reinstatement timeline. This is shaping up as Washington's most forceful commercial AI intervention yet — and it will set precedent for how future vulnerabilities are handled.

**OpenAI acquires Ona for persistent Codex agents** — OpenAI announced it's acquiring Ona, a startup that provides secure cloud environments for AI agents. Ona already supports 2 million developers working in persistent, reproducible cloud environments — and this gives Codex a place to work when you close your laptop. Context matters: Codex usage jumped 400% this year to over 5 million weekly users, and the most valuable work is shifting from minutes-long sessions to multi-day projects. Ona's technology lets agents access tools, systems, and scoped credentials inside a customer's own cloud, with full logging and governance. Undisclosed terms, subject to regulatory approval. This is OpenAI moving from "agent runs until session ends" to "agent runs until the job is done" — a fundamentally different model for agentic production workflows.

**OpenAI opens Oracle Cloud access to Codex and frontier models** — Also last week, OpenAI and Oracle announced that OCI customers can apply Universal Credits toward OpenAI models and Codex directly. No separate purchasing path needed — it's plumbing, but important plumbing. Enterprise buyers already committed to Oracle can now tap frontier models through existing cloud spend. Combined with the Ona acquisition, OpenAI is building the full stack: the model, the persistent execution environment, and now the enterprise procurement path. This puts pressure on cloud-agnostic agent platforms that rely on customers setting up separate API accounts. For enterprises already deep in Oracle, this is one less reason to look elsewhere.

That's the briefing for today. Have a great Sunday, Rich.