Tracks

Which track should I choose?

If your primary contribution is a formal model, analysis, or theoretical result, the Theory track is likely the best fit. If your primary contribution is a system, deployment experience, or operational insight, the Practice track is likely the better choice.

Many submissions naturally combine elements of both. In such cases, authors should choose the track that best reflects the main contribution they want reviewers to focus on.

Theory Track (12 pages)

The Theory track welcomes work that develops conceptual, mathematical, or algorithmic foundations for understanding and improving alignment in AI systems.

Submissions may study alignment from a variety of perspectives, including feedback dynamics, stability, robustness, guarantees, and tradeoffs that arise when systems operate over time.

Example topics include (but are not limited to):

• Formal models of feedback, control, and adaptation
• Stability and convergence of human-in-the-loop systems
• Analysis of alignment tradeoffs (e.g., safety, utility, privacy)
• Verification, auditing, or specification of aligned behavior
• Multi-agent or distributed alignment dynamics

Theory papers may include experiments when helpful, but strong theoretical contributions do not require large-scale empirical evaluation.

Practice Track (10 pages)

The Practice track welcomes papers grounded in real systems, deployments, and engineering experience. This includes systems that have been built, deployed, tested, or operated under real constraints.

Practice papers are encouraged to discuss not only what worked, but also limitations, tradeoffs, and lessons learned.

Example topics include (but are not limited to):

• Monitoring and auditing alignment after deployment
• Feedback and override mechanisms in production systems
• Privacy-preserving telemetry and observability
• Safety and control mechanisms for agents and tools
• Operational case studies and experience reports

Code or artifacts may be included when useful, using an anonymous repository during review.

What AlignSys is looking for

Across both tracks, AlignSys values work that connects alignment to system behavior over time. Submissions are especially encouraged to make assumptions explicit and to discuss how ideas interact with real-world constraints.

If you are unsure whether your work is a good fit, you are welcome to contact hello@alignsys.org with a brief abstract.