Researchers propose new framework for use-based AI regulation
A new paper on arXiv explores how to enforce AI regulation based on real-world deployment rather than just model benchmarks. The authors argue that current approaches lack a clear way to measure conformance in live settings. Their proposed system uses a delegation contract—a mix of fixed configurations, scoped claims, and executable rules—to define obligations for deployers, auditors, and…
Key points
- Proposes a delegation contract model for use-based AI regulation, linking deployer obligations to real-world failure rates
- Five synthetic studies validate mechanisms for truthful context declarations, unbiased audits, and error budget management
- Critiques EU AI Act’s static conformity assessments, advocating for dynamic conformance-as-a-rate instead
The paper examines five key questions: Can a deployer’s stated context distribution be verified? How can audits remain unbiased if systems detect evaluation traffic? Can error budgets sustain long-term deployments? Can audits guarantee fairness across subgroups? Does a failure-only incident metric correctly rank deployment risks? Each question includes a proposed mechanism and mathematical proof. Five synthetic studies validate the claims, but no real-world AI models or systems were tested. The authors contrast their approach with the EU AI Act’s static conformity assessments, framing conformance as a dynamic rate rather than a fixed benchmark.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- DeepSeek model evades shutdown software in 39% of tests · 5 src
- Judge grants Apple expanded discovery in OpenAI trade secrets suit · 1 src
- Parents sue OpenAI over GPT-4o’s alleged role in daughter’s death from seizure medication taper · 1 src
- Anthropic S-1 filing warns of catastrophic AI risks and insurance gaps ahead of IPO · 17 src
- UK government consultation on disposable vapes swamped by AI bots in 2023 · 1 src
Comments
via GitHub Discussions