900 billion valuation + recursive upgrades + national-level destruction—this script sounds more and more like a prequel to "The Terminator."

According to reports, Anthropic co-founder Jack Clark warned at Oxford that AI could achieve recursive self-improvement around 2028, capable of upgrading itself without human intervention. Taking the internal model Mythos as an example, it demonstrates the boundary of losing control: its power is comparable to a national-level cyber weapon, and due to its destructive potential, it has been indefinitely banned from public release, with only a few authorized organizations allowed to fix vulnerabilities. At the London Developer Conference, Anthropic also revealed its commercial ambitions, with rumors valuing the company at $900 billion, and Claude Code founder Boris Cherny promoting AI programming tools on-site.