AI Transparency Days — Research Hackathon 2026
Scientific research instead of MVPs.
Research frame
“Loss of control through autonomous agents”
Dates
20–22 November 2026
Venue
Fürth · Nuremberg region
Participants
Up to 100 researchers & students
Teams
3–5 people or solo sign-up
01Executive Summary & Vision
COAI deliberately breaks with the product-driven hackathon format. Instead of immature prototypes or MVPs, the focus is on substantial, open research questions.
The three-day sprint aims at robust scientific starting points: precise hypotheses, sound methodology, well-founded arguments or first experimental data. Promising approaches are to continue in a structured mentoring programme towards publication (workshop or conference papers).
Tech & policy on equal footing
Code is not a prerequisite. Rigorous governance, legal and policy analyses (e.g. regulatory scenarios, gap analyses) have the same methodological standing as technical implementations.
02The Research Frame
No isolated tracks in 2026: every team works under one shared, highly topical frame, viewed through three established perspectives.
Detect
Early-warning systems, metrics and anomaly detection for drift or goal deviation in agentic systems.
Understand
Systemic root-cause analysis, interaction patterns and false assumptions in the oversight of agents.
Control
Technical and regulatory safeguards: guardrails, emergency shutdown, intervention architectures.
03Tooling & Formats
Agent-Collabs Sandbox · optional
A modular multi-agent test environment for experimental work: one isolated instance per team, with example scenarios such as a trojan or sleeper within a collective.
Recommended output
Structured abstract
Max. 600 words in six sections, from scenario and thesis via method and result to outlook and references; appendix for code or logs.
Adequate evidence
Run logs, code repositories, source collections or coherent legal-conceptual argumentation.
Pitch
6 slides · 5-minute talk · 3-minute expert Q&A.
Open research · not judged
Open Stage
A jury-free presentation stage for innovative work outside the official research frame.
04Interdisciplinary Jury & Evaluation Rubric
The jury spans technology, policy, mechanistic interpretability and AI safety. Scientific quality is assessed with one rubric:
| Criterion | Weight | Core question |
|---|---|---|
| Contribution to the theme | 25 % | Is a real mechanism of loss of control addressed, with a robust gain in insight? |
| Method & reasoning | 25 % | Is the approach adequate and transparently documented, with its limitations reflected? |
| Evidence | 20 % | Are the claims backed by valid data, runs, cases or sources? |
| Scientific novelty | 15 % | Does the work go significantly beyond known demos and standard reviews? |
| Science communication | 15 % | Are question and results conveyed clearly and precisely across disciplines? |
Prizes are awarded to the winning teams. Judging criteria follow the rubric above; the prizes themselves will be announced closer to the event.
Pre-register
Places are limited to around 100 participants. Leave your name and email and we will contact you as soon as registration opens.
Contact
COAI gGmbH · Josephsplatz 8, 90403 Nuremberg, Germany
AI Transparency Days 2026 · hosted by COAI · Fürth · 20–22 Nov 2026 · status September 2026