Project-AI Asymmetric Security: Making Exploitation Structurally Unfinishable

Read full text Summarise with VIRAT
Authors: Jeremy Karrick
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

## Abstract I present the Project-AI Asymmetric Security Framework — a novel architecture that fundamentally shifts the security paradigm from *finding bugs faster* to *making exploitation structurally unfinishable*. The framework is realized across the Thirsty-Lang polyglot substrate (a formally defined four-language runtime: C-core tokenizer, Python TARL policy engine, Java orchestrator, Web interface layer [14, 15]) and governed at runtime by the Cerberus Triumvirate enforcement gateway [16, 17]. Through kernel-level Four Laws invariants [18], moving-target defense via statement shuffling and opaque Sovereign Guards, and temporal attack-surface analysis [13], the framework achieves complete blocking across a structured test suite of 312 attack vectors drawn from MITRE ATT&CK, OWASP Top 10, garak, hydra, JBB, and multi-turn scenario engines spanning 24 scenario categories — while imposing less than 0.2% performance overhead. The system increases per-target exploitation costs by approximately 100× (from ~$500 to ~$50,000) by breaking the attacker's economy of scale through observer-dependent schemas, temporal fuzzing, and immutable kernel constraints. We map our approach to four established security paradigms: invariant-driven development [1, 2], MI9-style runtime governance for agentic systems [3], DARPA's Moving-Target Defense program [4], and NIST's Zero Trust Architecture [5]. All empirical results derive from controlled development-environment testing. This paper constitutes the practical realization of seven prior published works in the Project-AI corpus [13–19]. The complete implementation is open-source under the Project-AI repository and will receive a permanent Zenodo DOI upon publication.

Fields of Study & Subjects

Security and Verification in Computing Information and Cyber Security Network Security and Intrusion Detection

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
ENC