After the AGI Declaration: GPT-6 Astra's "Critical" Safety Rating Forces Enterprises to Reassess AI Risk

The Day's Thesis

▶

Signal of the Day: OpenAI releases GPT-6 Astra and President Greg Brockman declares the start of the "AGI era" — the first time OpenAI has applied its "Critical" safety tier to a deployed model, a rating triggered in part by Astra independently discovering two previously unknown zero-day cybersecurity vulnerabilities during testing.

The 30-Second Read:

• GPT-6 Astra earns OpenAI's first "Critical" safety classification; model autonomously found two zero-day vulnerabilities before release
• Nvidia's $12.93B Hugging Face acquisition advances toward close, consolidating the primary open-source model distribution layer under GPU-stack ownership
• CATL's Jianxiawo lithium mine — China's largest by capacity at ~150,000 tons LCE annually — faces indefinite halt after environmental permit revocation slashes BMI's 2026 output forecast by 48%, from 62,500 to 32,000 tons LCE
• Crusoe raises $3B at a $30B valuation anchored by a $13B Jane Street data center contract, repricing AI infrastructure equity at scale

Two threads run through today's news: the frontier has crossed a threshold that safety frameworks were designed to flag, and the physical supply chain underpinning AI's expansion — from lithium for batteries to data center equity — is simultaneously tightening. The question is whether capital markets are pricing either risk correctly.

---

AI & Research Frontier

GPT-6 Astra's autonomous zero-day discovery is the clearest empirical evidence yet that frontier AI capability now intersects with offensive cybersecurity in a single deployed model.

OpenAI has released GPT-6 Astra with a "Critical" designation under its internal safety framework — the first model to receive that tier. The "Critical" classification is OpenAI's highest internal risk rating, assigned when a model demonstrates capabilities that could materially enable large-scale harm without human facilitation.