Self-Verifying AI Systems

Research on incomplete specifications, incorrect assumptions, and the burden of checking whether AI systems have met a user’s intent.

How can AI systems verify and improve their own work?

The central problem is the gap between producing a plausible answer and establishing that the system satisfied the user’s intent. When requirements are incomplete, systems may infer the wrong constraints or success criteria, leaving users to discover and correct those assumptions after the fact.

Methods and questions

automated validation
test generation
invariant checking
self-debugging
confidence and stopping criteria

Projects

Self-Verifying AI Systems

When specifications are incomplete, AI systems may make incorrect assumptions and leave users to discover the failures after the fact.

Publications

2026

BEGIN AI TRANSACTION: Semantic Isolation for Durable AI Workflows

Barzan Mozafari

arXiv 2026

Paper ↗Project ↗

Cite

@misc{83622372-dbac-4149-9b64-a2489c39f997,
  title = {BEGIN AI TRANSACTION: Semantic Isolation for Durable AI Workflows},
  author = {Barzan Mozafari},
  year = {2026}
}