Multiple "AI godfathers" have jointly warned of a runaway "intelligence explosion" (Guardian); almost simultaneously, Anthropic and OpenAI executives made a rare joint call for independent oversight of self-improving AI, with OpenAI joining Anthropic's earlier proposal for an international AI watchdog.

Why now

Because RSI (recursive self-improvement) is no longer a paper concept. This month Alibaba disclosed Qwen3.8-Max completing 33 rounds of zero-human self-iteration; agents now build training pipelines, construct data and fix defects autonomously. As models begin improving models, the oversight window is narrowing — that is the technical backdrop.

📊 This week's multi-source signals (Google News)

· Guardian: AI godfathers warn of runaway intelligence explosion;

· Yahoo Finance: Anthropic/OpenAI execs urge oversight of self-improving AI;

· CNN/CBS: public "AI doom" anxiety enters mainstream discourse;

· NYT: governments increasingly left behind.

A rational take

Panic doesn't help, but three levers deserve immediate action: interruptible kill-switches for self-improvement, independent red-team audits of training and deployment, and tiered disclosure of RSI capabilities. NineZenith keeps human confirmation gates and rollback in Zenith-Act's continuous training — every step of self-evolution should leave an interface to hit the brakes.

(Facts aggregated from Guardian, Yahoo Finance, CNN, NYT)