Zenodo ยท Jun 2026

Remarkable Convergence Anthropic's Safeguards and the Governance Twin

Reads the June 2026 dispute over whether Claude Fable 5 had been jailbroken through the distinction between control and governance. Anthropic built a safeguard whose heaviest layer sits outside the model, where the conversation cannot talk its way past it, but what it built is a gate that asks whether a single action is allowed, not a governing plane that holds an evolving standard and judges the trajectory. Applies the two cuts of Structure Not Content, and Trajectory Not Point to one public case, credits the separation, and names the steps from a very good gate to a governance plane, including what the operator who carries the liability inherits when it cannot set, evolve, or own the runtime standard.

Concepts

  • Governance Twin
  • Runtime Governance
  • Agentic AI

Keywords

  • Runtime
  • control
  • recoverability
  • admissibility
  • Agentic AI
  • Security

View on Zenodo → (opens in new tab)

← Back to all publications

Govern what you deploy.