What it looks at
It maps Anthropic’s Responsible Scaling Policy, OpenAI’s Preparedness Framework and its compliance framework, Google DeepMind’s Frontier Safety Framework, and Anthropic’s 2026 risk reports, incident investigations and monitoring statistics onto the four components and twelve elements of the SMS in ICAO Annex 19. It uses the documents public on 3 November 2026.
What it found
- There is more SMS in the documents than the headline frameworks suggest: hazard identification and risk acceptance throughout, incident response plans, and change management and audits for security.
- The gaps are specific. No developer publishes indicators with targets, alert levels and trends for its safety controls, such as monitors and classifiers. None documents a change process for those controls’ settings with a hazard assessment. None publishes an occurrence taxonomy or measures how long detection takes, and training for safety-assurance roles is not described.
- Anthropic’s oversight metrics (coverage, review latency and escalation rate) are safety performance indicators in all but name.
- It proposes twelve AI safety performance indicators for testing.
Why it matters
The gaps are the everyday tools of a safety office: indicators with alert levels, management of change and occurrence reporting. Aviation already runs them every day.
About this paper
This is a working paper, version v1.1 of 4 November 2026. It has not been peer reviewed. Every reference was checked against its source, and the paper was audited by a reviewer that did not write it. As the paper says, the literature search, drafting and checking were done with AI assistance under the author’s direction, and the author is responsible for the text. Its sources, audit trail and, where there is one, its code and data are on GitHub.
