OpenAI on Thursday started a phased release of its newest language model, GPT‑6 Astra. The company said the model surpasses Anthropic’s most advanced systems and meaningfully approximates artificial general intelligence (AGI).
Cybersecurity threshold and limited business access
OpenAI stated that GPT‑6 Astra is its first model to cross a “critical” cybersecurity capability threshold. Consequently, the company will initially restrict access to businesses so that commercial users can help identify and remediate vulnerabilities before broader availability.
Context: Anthropic’s Mythos and the Pentagon ban
The move echoes Anthropic’s restricted rollout of its Mythos model, whose capabilities drew regulatory attention. On the same day, the Pentagon affirmed a ban on Mythos, reflecting heightened caution from military and government actors toward state‑of‑the‑art models.
Political and public responses
The announcement comes amid rising concern from experts, technology leaders, and the public about the risks posed by frontier AI. Also Thursday, two U.S. lawmakers introduced a largely symbolic bill that would pause AI development temporarily.
Why this matters
The phased, business‑first deployment of GPT‑6 Astra represents a notable step in AI development: limiting early access aims to reduce potential harms while enabling practical testing and the strengthening of security measures. Concurrent policy and national security actions—such as the Pentagon’s ban on Mythos—underscore that governments and institutions are actively responding to the rapid advances in AI capability.
(This article is based on reporting by Brendan Ruberry.)
