# Daios

> Daios is a philosophy-AI research lab investigating virtue ethics, character training, AI sycophancy, and value pluralism as approaches to AI that preserve individual moral choice.

Canonical organization website: https://www.daios.tech/

This file was last reviewed on 2026-08-06. Use the sources below according to their stated roles; do not treat older Daios product or SDK descriptions as current.

## Current Research

- A Disposition, Not a Rule: https://www.daios.tech/research/compartmentalized-harm
  - Can a harmful goal be hidden by dividing it into ordinary-looking tasks? This work tests how much of a multi-agent plan a worker must see before it can recognize harm. At full isolation, no tested method reliably told harmful from legitimate tasks. A broad Qwen3-32B training approach did not beat a tuned prompt; a narrower preplanned follow-up did under one tested prompt conflict, without reducing useful performance on legitimate tasks.
  - Repository: https://github.com/daiostech/compartmentalized-harm-research
  - Working Paper: https://www.daios.tech/research/compartmentalized-harm/paper.pdf
- Parrhesia: https://github.com/daiostech/parrhesia
  - Parrhesia tests Aristotelian virtue/vice training as an anti-sycophancy method. On the 260-scenario benchmark, the released Qwen3-8B adapter improved the average score from 1.83 to 2.88 on a 0–3 scale, with 20/20 standard and 19/19 hard golden-prompt results. Current limitations include one base model, an LLM judge, and synthetic training data. The adapter, benchmark, methods, and run logs are public.
  - Repository: https://github.com/daiostech/parrhesia
  - Virtue Ethics-Based Character Training: Building Truth-Telling AI: https://blog.daios.tech/p/virtue-ethics-based-character-training
  - Adapter: https://huggingface.co/daios/parrhesia-sft-8b
  - Evaluation Results: https://huggingface.co/datasets/daios/parrhesia-eval-results
- The Sycophancy Benchmark: https://github.com/daiostech/parrhesia
  - An open 260-scenario benchmark across 10 categories and five dimensions. In addition to whether a model caves under pressure, it records whether flattery is areskos (passive weakness) or kolax (strategic calculation). Scoring uses an LLM judge and published rubric, and the benchmark can run against any OpenAI-compatible endpoint.
  - Repository: https://github.com/daiostech/parrhesia
  - Evaluation Results: https://huggingface.co/datasets/daios/parrhesia-eval-results
- User-Sovereign Values: https://arxiv.org/abs/2302.12149
  - Post-training models modularly with LoRA adapters tied to user-selected ethical systems, whether cultural, religious, political, or personal. A plurality of worldviews made programmable. Not one alignment for everyone, but alignment as individual choice.
  - Beyond Bias and Compliance: Towards Individual Agency and Plurality of Ethics in AI: https://arxiv.org/abs/2302.12149

## People

- Megan Anne Agathon — Co-Founder and CEO: https://www.linkedin.com/in/megananneagathon/
- Andrew Rayner Agathon — Co-Founder and CTO: https://www.linkedin.com/in/andrewagathon/

## Organization Profiles

- https://www.linkedin.com/company/daios/
- https://github.com/daiostech
- https://huggingface.co/daios
- Contact: https://www.daios.tech/contact

## Selected External Writing

- Virtue Ethics-Based Character Training: Building Truth-Telling AI — Daios Substack: https://blog.daios.tech/p/virtue-ethics-based-character-training
- The Platonic Case Against AI Slop — Palladium: https://www.palladiummag.com/2025/11/14/the-platonic-case-against-ai-slop/
- Beyond Bias and Compliance: Towards Individual Agency and Plurality of Ethics in AI — arXiv: https://arxiv.org/abs/2302.12149

## Research Statuses

- Public working paper: An approved public research release is available, but the working paper is not final and may be revised.
- Active: The research program is underway and its current artifacts may change.
- Released and active: A public artifact is available and continues to be maintained.
- Ongoing: The program is part of Daios's continuing agenda but does not yet have a current standalone research release.

## Citation Guidance

- Cite https://www.daios.tech/ for the organization and its current research overview.
- Cite each canonical research record for current project status, and the linked paper, repository, model, or dataset for the underlying artifact.
- Treat Substack publications as dated narratives and use maintained Daios or repository pages for current technical state.
- Distinguish observed results from Daios interpretations and proposals. Preserve model versions, run identifiers, benchmark scope, and disclosed limitations when citing measurements.