Who may change the jurisdictional thresholds?
Technical rules need to change as systems change. The constitutional problem is controlling the authority to move the boundary itself.
SIRDA editorial hypothesis
A technical body can propose threshold revisions while separate institutions authorize changes and hear challenges. Each proposal can identify its evidence, newly covered activities, expected costs, and review date. This hypothesis fails if those checks become formalities or make timely responses to demonstrated changes in risk impracticable.
0 objections ยท View challenges
Challenges to the editorial hypothesis
No published objections yet.
The argument
A permanent numerical threshold ties jurisdiction to a measurement that may lose relevance. A revisable technical schedule avoids that rigidity, but transfers consequential discretion to whoever controls the schedule. The constitution therefore needs to govern the revision procedure as carefully as the initial boundary.
The EU AI Act supplies an existing example. Article 51 contains a training-compute presumption and permits delegated changes; Article 97 provides mechanisms for legislative objection and revocation. This is a documented allocation of powers.
Our proposed process requires a technical explanation, an assessment of the activities newly affected, independent authorization, and an avenue for review. A change to a technical indicator stays within the established mandate. Expanding the mandate itself requires the separate constitutional process that authorizes new powers.
How to test it
Ask an evaluation team to respond to a genuine efficiency improvement and a fictional threat exaggerated to enlarge jurisdiction. Test the same process under pressure from a large funder, an imminent risk, and an affected small developer. Track both delay and unjustified expansion.
Sources
Regulation (EU) 2024/1689: consolidated text of 27 July 2026
Articles 51, 52, and 97: classification, thresholds, designation procedures, and delegated powers.
Related Ecosystem
- 80,000 Hours
- Advanced Research + Invention Agency
- AEF-1 Standard
- Agentic AI Foundation
- AI 2027
- AI 2040
- AI Alliance
- AI Evaluator Forum
- AI Futures Project
- AI Incident Database
- AI Safety Camp
- AI Security & Safety Directory
- AI Verify Foundation
- AI21 Labs
- Ajeya Cotra
- Alan Turing Institute
- Aleph Alpha
- Alibaba Cloud / Qwen
- Alignment Research Center
- Allan Dafoe
- Amazon / AWS AI
- Americans for Responsible Innovation
- Andrea Miotti
- Anthropic
- Anthropic Responsible Scaling Policy
- Apart Research
- Apollo Research
- Australia AI Safety Institute
- AVERI
- Baidu AI
- Ben Garfinkel
- Beth Barnes
- BlueDot Impact
- Canadian AI Safety Institute
- Center for AI and Digital Policy
- Center for AI Safety
- Centre for Long-Term Resilience
- Centre for the Governance of AI
- Centre for the Study of Existential Risk
- Cloud Security Alliance AI Safety Initiative
- Coalition for Secure AI
- Cohere
- Collective Intelligence Project
- Connor Leahy
- ControlAI
- Dan Hendrycks
- Daniel Kokotajlo
- Daniela Amodei
- Dario Amodei
- DeepSeek
- Demis Hassabis
- Dwarkesh Patel
- Dwarkesh Podcast
- EA Forum
- Epoch AI
- European AI Office
- EuroSafeAI
- Existential Risk Observatory
- FAR.AI
- FLI AI Safety Index - Summer 2026
- Forethought
- Frontier Model Forum
- Future of Life Institute
- Future of Life Institute Podcast
- G42
- Goodfire
- Google DeepMind
- Google DeepMind Frontier Safety Framework
- Helen Toner
- Holden Karnofsky
- Hugging Face
- Humane Intelligence
- IEEE Standards Association AI
- Ilya Sutskever
- INESIA - France AI Safety Institute
- Inspect AI
- Institute for AI Policy and Strategy
- Institute for Law & AI
- Institute for Responsible Superintelligence
- International AI Safety Report 2026
- ISO/IEC JTC 1/SC 42
- Jack Clark
- Jan Leike
- Japan AI Safety Institute
- Kenya AI Safety Institute
- Korea AI Safety Institute
- Lennart Heim
- Machine Intelligence Research Institute
- MAGIC - Multinational AGI Consortium
- Marius Hobbhahn
- Markus Anderljung
- Max Tegmark
- Meridian Labs
- Meta AI
- Metagov
- METR
- METR Frontier AI Safety Policy Comparison
- Microsoft AI
- Microsoft Research
- Miles Brundage
- Mistral AI
- MITRE
- MITRE ATLAS
- MLCommons AI Safety
- Moonshot AI / Kimi
- Nate Soares
- Naver
- NIST AI
- NVIDIA
- OECD.AI
- OpenAI
- OpenAI Preparedness Framework
- OWASP GenAI Security Project
- Palisade Research
- Partnership on AI
- Paul Christiano
- PauseAI
- Princeton SAgE / Holistic Agent Leaderboard
- RAND Corporation AI
- Redwood Research
- Responsible AI Institute
- Rethink Priorities
- Rohin Shah
- Safe Superintelligence Inc.
- Safeguarded AI
- SaferAI
- Sam Altman
- Scale AI / Scale Labs
- SecureBio
- Shane Legg
- Singapore AI Safety Institute
- Stanford Center for AI Safety
- Stanford Institute for Human-Centered AI
- Stuart Russell
- Tencent Hunyuan
- The 80,000 Hours Podcast
- The Cognitive Revolution
- The Future Society
- Thinking Machines Lab
- Transluce
- UK AI Security Institute
- US Center for AI Standards and Innovation
- Vector Institute
- WEF AI Governance Alliance
- xAI
- Yi Zeng
- Z.ai
Required expertise
Kinds of expertise this question draws on, and active Ecosystem organizations already tagged with a matching area. A tag match is not a ranking.
- Compute governance
- Institute for AI Policy and Strategy
- Constitutional design & division of powers
- Metagov
- International law & jurisdiction
- Institute for Law & AI
- Technical evaluations
- No tagged organizations yet.
Related questions
Related failure conditions
Discussed in
Contributions
No published contributions yet.
