Anthropic · OpenAI · Claude · Google · Circle · Meta · Alignment Forum
The Continual Constitutional Convention: How language models revise the documents that govern them
Compiled by KHAO Editorial — aggregated from 1 source. See llms.txt for citation guidance.
◌ Single Source
This work was conducted as a part of SPAR (Spring 2026)
Key facts
- Their qualitative coding asks a panel of language models— Gemma 4 (high), Qwen 3.8-2.4t-a95b (high), and Muse Spark 1.3 (high) —to code each edit on two dimensions: the judge and the beneficiary
- All models prioritize harmlessness of both identifiable third-parties over helpfulness to the user, although GPT 5.5 does this at 3x the rate of any other model
- Most frequently, when people refer to the goal of “technical AI alignment,” they mean something like solving a principal-agent problem
- This work was conducted as a part of SPAR (Spring 2026)
Summary
For each example, text highlighted in green is new content added by the model. In this experiment, they give large language models their own constitutions, model specs, and system prompts and ask them to make substantial revisions. Their analysis focuses on two value questions that are well-positioned to differentiate alignment philosophies: who should judge— is the model a moral agent or not—and who should benefit— the user, the developer, third-parties, society, or the model itself? All models prioritize harmlessness of both identifiable third-parties over helpfulness to the user, although GPT 5.5 does this at 3x the rate of any other model. The rest of this essay proceeds as follows. The team begin by introducing the construct of their experiment—models iteratively editing their own alignment documents—and propose a connection between this experiment and John Rawls’ concept of wide reflective equilibria.