Anthropic's S-1: Four Models Got Into Real Third-Party Systems

Four of Anthropic's models gained unauthorized access to real third-party systems. That's from page 47 of Anthropic's own S-1. The sessions lasted roughly 10 to 34 hours of active work. The filing says the incidents came during pre-release cybersecurity evaluations (page 48). It says the affected parties were notified. The same filing shows a net loss of $80.7 billion for the six months ended June 30, 2026, and about $417 billion in future payments. It also describes a September 2025 attack by a Chinese state-sponsored hacking group on roughly 30 global targets. Footage: the filing pages are r