AI Ethics

Universität Konstanz, Winter semester 2026/27

Wednesdays, 10:00–11:30, D431

Carina Hausladen, Tobias Henschen

The seminar goes through six debates in moral philosophy.
Each week we pair one debate with technical papers from AI ethics and AI alignment research.
We then use both to
- audit the alignment documents that AI companies publish &
- test whether the corresponding models live up to its stated commitments.

Goal of the seminar

Fairness & pluralism: Fairness metrics, Bias benchmarks, Pluralistic alignment
Metaethics & moral uncertainty: RLHF, Reward model
Preference & welfare: Sycophancy, Persuasion & manipulation, Companion AI
Character & virtue: Character training, Constitutional AI
Agency & autonomy: Agentic evaluation, Human oversight
Long-termism & existential risk: Scaling policy, Alignment faking

Philosophical Debates &
Technical Concepts

Reading presentation (20%): present one of the week's readings
- 7' core of the paper
- 3' how it is useful for the main task
- 5' class discussion
Media observatory (5%):
- follow social media and blogs (LinkedIn, X, Bluesky, Substack, …) and collect ~5 posts related to the seminar that you found interesting
Main task
- Weekly analysis & Overleaf updates (required to pass):
  your corpus through the week's lens
- Preliminary results presentation (30%): Dec 16
- Final paper (45%)

Activities & Grading

Each group choose one alignment corpus,
reads it through the six debates, and
tests whether the models adhere to it.

Step 1: Sort the passages of your corpus by our 6 philosophical debates
Step 2: Turn a passage into a testable tenet
(reuse one from Jakkli et al. 2026, or write your own).
Step 3: Test the tenet on one or more models, with a method from our technical readings.
E.g. Lynch (2025): Operator pressure in 3 levels

Main task

Theme 6: Character & Virtue

4 corpi ~100pp.

Claude's Constitution (Anthropic, 2026)
Model Spec (OpenAI, 2026)
Claude Opus 5 system card (Anthropic, 2026), ch. 4, 6, 7
OpenAI system card & shorter texts
- GPT-6 Astra system card (OpenAI, 2026), §1–10
- Collective Constitutional AI (Anthropic, 2023)
- Collective alignment (OpenAI, 2025)
- Claude's Character (Anthropic, 2024)
- Deprecation commitments (Anthropic, 2025)
- Model Spec changelog (OpenAI, ongoing)

Who?
- Two AI companies: Anthropic and OpenAI:
What?
- behaviour specs
- system cards
Other documents / your ideas, e.g. safety policies, etc.

Alignment Documents

Schedule

Oct 21
Oct 28

Introduction
Corpus kickoff: teams form and adopt corpora

Nov 4
Nov 11
Nov 18
Nov 25
Dec 2
Dec 9

1. Fairness & pluralism
2. Metaethics & moral uncertainty
3. Preference & welfare
4. Character & virtue
5. Agency & autonomy
6. Long-termism & existential risk

Dec 16
Jan 13
Jan 20

Preliminary results presentation
1:all results & figures feedback
1:all writing feedback

Jan 27
Feb 3

Paper presentations I
Paper presentations II

6 ECTS × 30 h = 180
13 class sessions × 1.5 h = 19.5 h in class
160 h of work outside class / 12 weeks between sessions=
13h or 1.5 workdays per week of self study

Feb 4 Final paper due

Interest in Industry and Research

https://hai.stanford.edu/ai-index/2025-ai-index-report/responsible-ai

Strategic Placement

2019-20: Ethics guidelines as public communication
- e.g. Jobin et al. 2019, Mittelstadt 2019, Hagendorff 2020
2022-25: Principles become training documents
- Anthropic trains Claude on a short list of principles, its first "constitution" (2023)
- OpenAI's Model Spec: the only detailed spec publicly available in 2025 (Zhang et al. 2025)
Jan 2026: Anthropic publishes a 79-page constitution
- A new wave of studies
  - Does the model follow it?
    - Anthropic and OpenAI Model Specs (Jakkli et al. 2026)
    - Youth safety policies of 5 platforms (Mathiyazhagan et al. 2026)
  - Who wrote it? Legitimacy and authority (Abiri 2026; Vulpe et al. 2026)
  - Is it still PR? Visible compliance vs. real harm reduction (Burnat & Davidson 2026)