Anthropic CEO Dario Amodei said the company held a finished Mythos-class model from release for four months. It was limited to cyber-defense partners until June. Amodei cited other costly safety choices including biological-risk filters. He also pointed to support for California’s SB 53 and a 5% equity commitment to a public benefit fund. He conceded some criticism from Encode AI’s Nathan Calvin had merit. The exchange follows a draft federal AI safety bill Calvin criticized. Anthropic is preparing a November IPO targeting a $2 trillion valuation. It reported a first-half net loss above $80 billion on $16.3 billion revenue. Some investors now discuss a $1.5 trillion valuation instead. The company may spend up to $750 billion on computing capacity in coming years.

• 4-month delay on Mythos-class model
• $80B+ H1 2026 net loss
• $16.3B H1 2026 revenue
• $2T targeted IPO valuation