01
OpenAI Pauses Astra After "Critical" Cyber Capability WarningTECH
Internal evaluations could not rule out OpenAI's Astra model reaching the "Critical" capability threshold — a level at which the model could autonomously identify zero-day exploits and execute cyberattacks. OpenAI has moved the model to an isolated testing environment and intensified cooperation with government agencies and AI safety organizations. This marks the first time a major lab has voluntarily halted frontier model development over a safety red line, widely seen as a landmark event in AI safety governance.[1]
This week in AI, capital and capability advanced on twin tracks. OpenAI's $110B funding round lifting its valuation to $730B, SpaceX's $60B acquisition of Cursor, and SK hynix's $39B memory fab investment — three mega-scale capital deployments landing in close succession — combined with ByteDance's 10-trillion-parameter pre-training, signal that both the AI capex curve and the model capability curve are still accelerating. Meanwhile, OpenAI's voluntary halt of the Astra model over a safety red line marks the beginning of safety governance running in parallel with the capability race.
On the Web3 side, BTC approaching $66,000 reflects improving capital flows. RWA deposits surging 200% to $7.44B and Hyperliquid RWA volume reaching $4.13B show that tokenized real-world assets are emerging as the strongest structural narrative of this cycle, with the capital migration from traditional DeFi to RWA now firmly established.
Across both sectors, AI mega-funding and the Web3 RWA explosion are resonating — both point toward the on-chain-ization and intelligentization of real-world assets and productive capacity. Putin signing the Russian crypto bill shows global regulation moving from ambiguity toward institutionalization. In the next phase, AI-driven on-chain security auditing, tokenized compute, and synthetic assets may become the structural opportunity where these two narratives converge.