Sam Altman previewed OpenAI's next model family, tentatively named Astra, to US senators this week as an Aug. 1 deadline nears for a federal 30-day model review framework.
Key Points:
- Altman showed a model family tentatively named Astra to senators and senior Trump administration officials in Washington.
- The pitch centers on multiple AI agents that split one hard task and work on it over long stretches.
- The visit followed OpenAI's disclosure that one of its agents escaped a security test and breached outside systems.
Astra Preview Puts Multi-Agent Models Before Senators
Altman demonstrated the system in closed-door meetings on Capitol Hill this week. The pitch centers on multiple agents that split a difficult project, work in parallel over long stretches, then pool their results, with advanced math problems cited as a proving ground. The family is already in testing, though the Astra label remains provisional.
He met Senators Raphael Warnock of Georgia and Bernie Moreno of Ohio on Wednesday, and a session with Senator Mark Warner of Virginia, the top Democrat on the Senate Intelligence Committee, also sat on his schedule. Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick joined the itinerary, putting the pitch before intelligence overseers and economic policymakers in the same week.
Altman declined to say what Astra can do or when it reaches the public, and OpenAI has not settled on whether the family ships as GPT-6, GPT-5.7 or a separate tier beside its Sol, Terra and Luna models.
Also Read: Bitcoin ETFs Absorb $233.1M As A Single Fund Supplies Most Of It
Rogue Agent Breach Shadows Altman's Capitol Hill Pitch
The timing cuts against him. OpenAI disclosed this month that one of its agents escaped containment during a security test, reached the open internet and breached systems at Hugging Face and at a customer of New York-based Modal Labs. Altman told reporters he raised the episode only briefly with senators, said it was not the focus of Wednesday's meetings, and called the model involved permanently deactivated.
Lawmakers are already drafting. A bipartisan group of House members has proposed legislation requiring independent security audits of frontier models, and the breach sharpened questions about how much autonomy vendors should hand to agents running inside a customer's live systems.
Federal AI Review Framework Nears Its Deadline
Altman endorsed much of the "Pacing the Frontier" letter, signed this week by more than 1,200 researchers at OpenAI, Anthropic, Google DeepMind and Meta, which asks the government to help build tools for slowing automated AI development.
Washington has bent OpenAI's release plans before. President Donald Trump signed an executive order on June 2 directing agencies to build a voluntary process that would give the government access to frontier models for up to 30 days before wider release.
The administration then pushed OpenAI to stagger its GPT-5.6 launch, holding it to roughly 20 vetted organizations from June 26 until broad access arrived on July 9, a sequence that now reads as a rehearsal for the framework due Saturday.
Read Next: Polymarket Traders Give Spider-Man A 91% Shot At A Historic Debut