# ANOSY: Approximated Knowledge Synthesis with Refinement Types for Declassification

arXiv: https://arxiv.org/abs/2203.12069 · HTML: https://mlsanity.com/paper/2203.12069/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 22 Mar 2022.

## Built on by (most cited first, 1 of 1)

- [Confuse the Model, Control the Flow: Understanding and Mitigating Privacy Leakage from LLM Agents with Information Flow Control](https://mlsanity.com/paper/2609.14003/) (2026)

## Abstract (arXiv)

Non-interference is a popular way to enforce confidentiality of sensitive data. However, declassification of sensitive information is often needed in realistic applications but breaks non-interference. We present ANOSY, an approximate knowledge synthesizer for quantitative declassification policies. ANOSY uses refinement types to automatically construct machine checked over- and under-approximations of attacker knowledge for boolean queries on multi-integer secrets. It also provides an AnosyT monad to track the attacker knowledge over multiple declassification queries and checks for violations against user-specified policies in information flow control applications. We implement a prototype of ANOSY and show that it is precise and permissive: up to 14 declassification queries are permitted before a policy violation occurs using the powerset of intervals domain.