How should military autonomy fit within the mandate?
Military applications test the proposed boundary where secrecy and security interests are most consequential. The inquiry must explain both the coverage it claims and the risks a limited mandate leaves outside.
SIRDA editorial hypothesis
A capability-based mandate can assess catastrophic risk across civilian and military deployments using distinct access and review protocols. A civilian-first phase can be evaluated only with an explicit account of the risk left outside its coverage. This hypothesis fails if military relabeling defeats the common boundary or if claimed oversight depends on evidence access that participants will not provide.
0 objections ยท View challenges
Challenges to the editorial hypothesis
No published objections yet.
The argument
The scope decision has two costs. Including military activity creates difficult requirements for participation, evidence access, and protected review. A civilian-first arrangement leaves important activity outside its direct authority. Both belong in the same assessment of the institution's practical coverage.
The question here concerns catastrophic AI risk within military and intelligence applications. Operational command, battlefield decisions, and wider military policy remain separate issues. A scope proposal needs to identify precisely which development or deployment conditions trigger review and what information is necessary to establish them.
The inquiry will compare coverage models using the same scenarios. A protected review process must still allow meaningful challenge to the evidence. A narrower initial mandate must record its residual risks, dependence on other arrangements, and the conditions under which those omissions undermine its stated purpose. Labeling a gap is the beginning of that evaluation.
How to test it
Take a single frontier system through civilian research, dual-use procurement, and military designation while holding capability constant. Identify where review changes, which evidence remains available, and what prevents designation alone from defeating the threshold. Compare a civilian-first launch with broader but less verifiable coverage.
Related Ecosystem
- Center for a New American Security
- Center for Security and Emerging Technology
- CISA AI Security
- DoD Chief Digital and AI Office
- International Committee of the Red Cross (ICRC)
- NSA AI Security Center
- NVIDIA
- Stockholm International Peace Research Institute (SIPRI)
- Stop Killer Robots
- UN Institute for Disarmament Research (UNIDIR)
Related questions
Related failure conditions
Discussed in
Contributions
No published contributions yet.