June–August 2026
Authority by Design
Keep the power to judge, intervene and correct.
Memo 04 · GitFlash Memos · GitFlash · Retrospective thematic edition · Prepared 6 September 2026 · Published 7 September 2026
10 pages · approximately 12 minutes · Research cutoff 6 September 2026
The decision question
Can the person accountable for a decision still understand, challenge and stop what the system makes easy?
A decision memo on testing authority in an AI-assisted workflow instead of inferring control from a human signature.
Preview
AI can improve a decision while changing who can actually make it. A named approver may lack evidence, time or a working intervention.
This memo proposes a 90-day comparison of complete workflows, including error detection, intervention and correction under realistic pressure.
Contents
- Approval versus effective authority
- What policy and experiments establish
- Five tests of retained control
- Three divisions of work
- A 90-day comparative evaluation
Takeaways
- Compare the complete workflow with a credible alternative.
- Demonstrate that intervention changes the outcome before relying on it.
- Treat correction and appeal as part of delivery.
Sources
- 1. M-25-21: Accelerating Federal Use of AI through Innovation, Governance, and Public Trust
Office of Management and Budget · 3 April 2025
Agency-use policy with scope limits. High-impact status can apply despite human oversight; National Security Systems are excluded.
- 2. AI Risk Management Framework
National Institute of Standards and Technology · 26 January 2023; index accessed 6 September 2026
A voluntary framework for incorporating trustworthiness into AI design, development, use and evaluation. It is not a grant of legal authority.
- 3. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
National Institute of Standards and Technology · 26 July 2024
NIST AI 600-1. Identifies risks from human–AI configurations, including excessive reliance and inappropriate aversion.
- 4. When combinations of humans and AI are useful: A systematic review and meta-analysis
Michelle Vaccaro, Abdullah Almaatouq and Thomas Malone; Nature Human Behaviour · 28 October 2024
Original synthesis of experiments published in 2020–2023. Heterogeneous results do not establish the performance of a particular current workflow.
- 5. National Security Presidential Memorandum/NSPM-11
The White House · 5 June 2026
National-security AI directive combining accelerated adoption with assurance and accountable command. Implementation outcomes and subsequent guidance are not established here.