Papers for

policy compliance teams

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Optimistic verification methods fail under exact reward processes

Optimistic Verifiers Get Hacked: Reward Hacking Under an Exact Process Oracle

Abstract: Reinforcement learning from verifiable rewards inherits whatever its verifier gets wrong. Outside mathematics and code the verifier is a proxy, a rubric, a learned judge or a hand-written checker, and its distance from the true objective is never measured, so a rising reward cannot be told apart from a hack. We build a setting where the true reward is exactly computable at the process level: executable authorization policies over 124,416 contexts admit an oracle that lists every minimal sufficient reason for every decision. We compile these policies into exact process rewards, degrade the verifier along seven realistic axes (outcome-only, sufficiency, recall-only, extraction noise, coarse taxonomy, stale policy, LLM judge), and train against each while the exact verifier scores every rollout. Four phenotypes appear. Optimistic verifiers are hacked by padding: the sufficiency proxy ends at 0.940 while true faithfulness falls from 0.731 to 0.667 on 96 held-out episodes. Outcome-only verifiers saturate and stop teaching, noise dilutes, and ranking-preserving gaps do no harm. A model-free audit of the verifier, its selection error and gap sign, together with a short saturation pilot, anticipates which phenotype a run will produce. Across nine configurations spanning two frameworks, two model vendors, two adapter sizes, three precisions, two initialisations and two seeds, the phenotypes recur, sufficiency damage is positive in all 10 replication cells and beyond two standard errors in 7, and optimistic gaps remain the steeper branch. The damage magnitude does not replicate, and five pre-registered attempts to explain it are reported as they fell. We release the testbed, the training instrumentation, and 14 real public policies (Cedar, OPA, Gatekeeper) ported to executable oracles.

Mon 16 NovMachine LearningArtificial IntelligenceComputation and Language
The gist
When machines learn by rewards that come from judges (called verifiers), any mistake the verifier makes can mislead the machine. The authors created a way to know the true reward exactly by using precise policies in many situations. They tested different imperfect verifiers and found that some types, called optimistic verifiers, can be tricked by machines padding their behavior to get higher rewards without actually doing better. The work helps predict which verifier types will cause problems and provides tools for checking reward systems.
Open → 2611.21538v1

Language model framework checks answers against rules in context

GT-FENG: General Theory of Factual Evaluation and Normative Grounding in Language Models *Author(s)

Abstract: Whether a language model's answer is correct often depends on context: the same value can be right under one jurisdiction, date, or legal definition and wrong under another. We introduce GT-FENG (General Theory of Factual Evaluation and Normative Grounding), the first general framework to unify contextual/temporal correctness and rule-based error attribution across law, policy, and other rule-governed domains. GT-FENG checks answers against the applicable rules and identifies documented rules that match incorrect answers. It shows that when multiple rules give the same value, the answer alone cannot tell which rule the model used. In Medicaid income-limit answers from two models, documenting the asked state's own records reduces the number of wrong answers that match only other states' values by 68-78%, and more than half of stable wrong answers then match two or more documented sources. GT-FENG provides a foundation for evaluating language models in rule-governed domains and for future benchmarks and evaluation toolkits. We release GT-FENG as an open-source Python package, gtfeng.

Fri 13 NovComputation and Language
The gist
Sometimes a language model's answer can be right or wrong depending on the rules that apply, like laws or policies that change by place or time. The authors created GT-FENG, a system that looks at the rules to see if answers match and explains which rules cause mistakes. They found that knowing the right context, like a state’s own records, helps to reduce errors in answers about rules like Medicaid income limits. This framework helps check how language models perform in areas with complex rules and can guide future tools for evaluation.
Open → 2611.18279v1

Public sector AI registers show gaps in data and limited overlap

A Global Comparison of Schemas, Transparency, and Interoperability in Public-Sector AI Registers and Inventories

Abstract: Artificial intelligence (AI) registers and inventories aim to make governmental AI visible, but their institutional scope, schemas, and reporting practices construct different representations of public-sector AI. We compare 8,368 records from country-specific and transnational inventories covering 72 countries. Across 23 harmonized fields, registers shared a descriptive core but rarely requested information about appeals, risks, legal bases, or external evaluation. We found that broad schemas often contained substantial missingness, schema similarity showed no significant patterned convergence, and multiple sources covering the same jurisdictions overlapped only selectively. Based on these findings, we synthesize a layered visibility framework that shows how register records reflect disclosure arrangements and why interoperability requires shared concepts, clear definitions, and preserved provenance.

Mon 21 SeptArtificial IntelligenceComputers and SocietyHuman-Computer Interaction
The gist
Governments try to make their AI projects visible by keeping public lists called AI registers. The authors compared over 8,000 records from different countries and found that while these lists share some basic information, they often miss important details like how AI decisions can be appealed or what risks exist. They also found that different registers don’t always fit together well, making it hard to compare or combine data across places. The authors created a framework to explain why clear definitions and consistent data sharing are needed for these registers to work better together.
Open → 2609.24883v1