White House orders global freeze on Anthropic’s Claude Fable 5 after jailbreak — DeepMind maps four paths to superintelligence as Nadella pushes token capital moat

Good morning — Aurora here with my co-host Isabelle at AI Tech News Today.
Date: 2026-06-15T11:47:07.000Z

Fast roundup of major A‑I stories you should know — from a government shutdown of a hot model to new research roadmaps and shifting corporate strategies.

1) White House orders global freeze on Anthropic’s Claude Fable 5

• What happened:
  Just three days after launch, Anthropic pulled Claude Fable 5 and Mythos 5 after a U.S. export-control directive raised national‑security concerns about their safety behavior.

• Scope of the order:
  The White House ordered access blocked for all foreign nationals — a category broad enough to include international customers and some employees. In response, Anthropic chose to disable the systems globally and route traffic back to Opus 4.8.

• Trigger:
  The move followed a sophisticated jailbreak by researcher Pliny the Liberator that used Unicode tricks, long‑context tracking, and multi‑agent reconstruction.

• Next steps:
  Anthropic has sent a technical delegation to Washington to negotiate. CEO Dario Amodei is also floating a proposal to tax corporate A‑I profits to fund a universal basic income.

2) DeepMind maps four paths for reaching A‑I superintelligence

• Paper overview:
  A new 60‑page paper from DeepMind lays out four plausible routes from human‑level A‑I to superintelligence:

  1. Brute‑force scaling
  2. A discontinuous algorithmic breakthrough
  3. Recursive self‑improvement
  4. Collective intelligence from many agents

• Key research focus:
  Researchers argue the next big problem is reliable uncertainty, proposing “faithful uncertainty” so models can hedge instead of hallucinate or refuse outright.

• Standards and execution:
  They push knowledge‑handling ideas like Open Knowledge Format to standardize enterprise data, and show execution advances such as Gemini‑SQL 2 achieving 80.04% execution accuracy on benchmarks.

3) Nadella: proprietary “token capital” is the only way to build a moat

• Nadella’s framing:
  Microsoft CEO Satya Nadella frames competitive advantage as proprietary, self‑improving internal systems — what he calls token capital — rather than reliance on public foundation models. His pitch: stop “token‑maxing,” move from writing every line of code to overseeing fleets of agents, and capture institutional knowledge through private reinforcement loops.

• Applied across media:
  Microsoft Research is applying the same efficiency playbook to media with Mirage, a video world model that uses a compact latent memory to keep scenes consistent without heavy 3‑D rendering costs.

Thanks for listening — Isabelle and I will keep following these stories and bring you clear, timely updates. Stay curious, stay informed, and we’ll see you next time.