The Right for AI to Remain Silent

In The Right for AI to Remain Silent, Formic AI introduces Formic Engine, an embedded deterministic middleware stack, comprising a proprietary Thesaurus Graph, Lossless Context Serialization, and Predicate Gating, built to eliminate generative hallucinations and the manual "verification tax" in regulated, high-liability deployments. Rather than coupling lossy vector chunking with unconstrained autoregressive generation, Formic Engine compiles enterprise corpora into a schema-enforced relational Abstract Syntax Tree (AST), pins every propositional claim to an immutable sentence-level provenance graph, and mathematically verifies predicate satisfiability prior to synthesis, executing an explicit "Not in Context" abstention whenever source records cannot warrant an answer. Evaluated across 732 queries on two adversarial legal benchmarks, CanLegalRAGBench (532 Canadian case-law queries) and the Victorian Legal RAG Benchmark (200 dual-pipeline criminal law evaluations), the system achieved a 0.0% audit-verified hallucination rate across all 559 generated responses while deterministically abstaining on 173 ungrounded or corpus-absent inquiries. Formic Engine delivered a 0.7811 full-benchmark MRR on CanLegalRAGBench (+18.2% relative lift over iterative state-of-the-art baselines, and 0.9508 MRR with 95.08% Hit@1 on warranted queries), reached 96.0% auditor-adjusted accuracy on the Victorian benchmark (outperforming GPT-5.2 and Gemini 3.1 Pro baselines), and demonstrated near-total Foundation Model Invariance across distinct frontier generators, showing that deterministic infrastructure can commoditize downstream LLMs, slash inference compute costs, and establish a verifiable engineering standard for mission-critical AI, with all raw benchmark data and audit logs open-sourced for independent inspection.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24288408
Primary Topic
Artificial Intelligence in Law
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

The Right for AI to Remain Silent

Varun Ranganathan, Souvik Kundu, Daniel Escott
3 links
Zenodo (CERN European Organization for Nuclear Research)
Artificial Intelligence in Law
article

The Right for AI to Remain Silent

Varun Ranganathan, Souvik Kundu, Daniel Escott
article en

Abstract

In The Right for AI to Remain Silent, Formic AI introduces Formic Engine, an embedded deterministic middleware stack, comprising a proprietary Thesaurus Graph, Lossless Context Serialization, and Predicate Gating, built to eliminate generative hallucinations and the manual "verification tax" in regulated, high-liability deployments. Rather than coupling lossy vector chunking with unconstrained autoregressive generation, Formic Engine compiles enterprise corpora into a schema-enforced relational Abstract Syntax Tree (AST), pins every propositional claim to an immutable sentence-level provenance graph, and mathematically verifies predicate satisfiability prior to synthesis, executing an explicit "Not in Context" abstention whenever source records cannot warrant an answer. Evaluated across 732 queries on two adversarial legal benchmarks, CanLegalRAGBench (532 Canadian case-law queries) and the Victorian Legal RAG Benchmark (200 dual-pipeline criminal law evaluations), the system achieved a 0.0% audit-verified hallucination rate across all 559 generated responses while deterministically abstaining on 173 ungrounded or corpus-absent inquiries. Formic Engine delivered a 0.7811 full-benchmark MRR on CanLegalRAGBench (+18.2% relative lift over iterative state-of-the-art baselines, and 0.9508 MRR with 95.08% Hit@1 on warranted queries), reached 96.0% auditor-adjusted accuracy on the Victorian benchmark (outperforming GPT-5.2 and Gemini 3.1 Pro baselines), and demonstrated near-total Foundation Model Invariance across distinct frontier generators, showing that deterministic infrastructure can commoditize downstream LLMs, slash inference compute costs, and establish a verifiable engineering standard for mission-critical AI, with all raw benchmark data and audit logs open-sourced for independent inspection.

Zenodo (CERN European Organization for Nuclear Research)
Peace, justice, and strong institutions
Artificial Intelligence in Law
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon