Causal AI
Developing frameworks for causal inference in complex systems, including the dotcausal binary knowledge graph format and the foss-generator training data pipeline.
Independent Researcher
David Tom
David Tom
Foss
Causal AI · Post-Quantum Cryptography · Sovereign AI · Defence AI
Görlitz, Germany
Independent researcher at the intersection of formal logic, cryptographic security, and artificial intelligence. IEEE member, published author, European Parliament accredited.
Background
About David
David Tom Foss is an independent researcher based in Görlitz, Germany, working at the intersection of causal inference, post-quantum cryptography, and artificial intelligence. After nine years of service with the German Armed Forces (Bundeswehr), he transitioned into independent research, bringing a unique combination of operational military experience and academic rigour.
His primary research focus is the development of the Compression-Adjusted Structural Integrity (CASI) metric—a black-box statistical framework for validating cryptographic implementations. This methodology has been peer-reviewed and accepted at IEEE ICECET 2026 in Rome, where he will present two papers on post-quantum cryptography. Two additional ICECET papers are currently under peer review. A separate technical report on IBM z/OS mainframe security, published by FOSS Intelligence Holdings, documents 50 findings across 8 security domains.
Beyond cryptography, Foss is developing foundational tools for causal AI: the dotcausal binary knowledge graph format and the foss-generator synthetic training data pipeline. These address a fundamental problem in AI-assisted discovery—the tension between LLM hallucination and structured reasoning.
He is a registered lobbyist in the Deutscher Bundestag Lobbyregister, holds European Parliament accreditation, and is the founder of FOSS Intelligence Holdings Ltd. (UK). His work spans academic publication, open-source software development, and 15+ DPMA patent applications.
15+
Patent Applications
4
IEEE ICECET Papers
50
Security Findings (IBM)
13
Publications
2.5K+
Total Reads
7
Open Source Repos
Focus Areas
Research Areas
Post-Quantum Cryptography
Black-box statistical analysis of NIST PQC standards (ML-KEM, ML-DSA, HQC) using the CASI metric. IEEE ICECET 2026 peer-reviewed. First compression isolation of distributional signatures in PQC ciphertext.
Sovereign AI
Analyzing European AI sovereignty, GDPR-compliant AI infrastructure, and digital autonomy.
Defence AI
Bridging military domain expertise with cutting-edge AI research for national security applications.
Coined Terms & Frameworks
Frameworks & Systems
Original frameworks, systems, and tools conceived and developed by David Tom Foss.
CASI
IEEE Peer-Reviewed
Compression-Adjusted Structural Integrity
Black-box statistical framework for post-quantum cryptographic validation. Isolates distributional signatures in ciphertext by removing compression artefacts. Peer-reviewed and accepted at IEEE ICECET 2026, Rome.
KONPEKI
Live
Maritime Surveillance AI
Maritime surveillance AI system that identified sanctioned vessels, prompting a formal Norwegian government response. AI-powered vessel tracking and sanctions enforcement monitoring.
PARRHESIA
Research
AI Model Forensic Integrity System
AI model forensic integrity verification system. Detects and analyses model tampering, training data contamination, and adversarial manipulation in deployed AI systems.
GOTTFORMEL
Research
Autonomous Physics Discovery System
Autonomous physics discovery system. Applies causal inference and distributional analysis to identify novel physical relationships from experimental data without prior hypotheses.
live-casi
Python · PyPI
Black-Box Encryption Validation Tool
Python package implementing the CASI metric for black-box validation of cryptographic implementations. Covers 8 cipher families including NIST PQC standards ML-KEM, ML-DSA, and HQC.
dotcausal
Python · PyPI
Binary Knowledge Graph Format
Binary knowledge graph format with embedded 3-pass deterministic inference. Solves the core problem in AI-assisted discovery: LLMs hallucinate, databases don't reason. The .causal format bridges both.
Academic Output
Publications
Top 18%
SSRN Author Rank — worldwide
2,500+ Reads · 178 Downloads · 13 Papers · 22 Zenodo Records
2026
Accepted · Peer-Reviewed · IEEE ICECET 2026
Causal Graph Topology for Automated Security Margin Analysis and Blind Cipher Identification
2026
Accepted · Peer-Reviewed · IEEE ICECET 2026
Compression Isolation of Distributional Signatures in NIST Post-Quantum Ciphertext
Scheduled: Computing Methodology eJournal (Mar 4, 2026) · Cybersecurity, Privacy & Networks eJournal (Sep 29, 2026)
2026
Under Review · IEEE ICECET 2026
Persistent Cross-Round Carry Leakage in ARX Ciphers: Detection, Prediction, and Topological Classification
2026
Under Review · IEEE ICECET 2026
Deterministic Validation for Reliable LLM-Based Causal Knowledge Extraction
Scheduled: AI Intelligence eJournal (Jul 15, 2026) · Cybersecurity, Privacy & Networks (Sep 10, 2026)
2026
Technical Report · FOSS Intelligence Holdings
Cryptographic Security Assessment of IBM z/OS Mainframe Infrastructure Using CASI Distributional Analysis
2025
2026
Distributed · eJournal
.causal Cryptanalysis: Black-Box Security Margins and Blind Cipher Identification
Scheduled: AI eJournal (Sep 29, 2026) · Computing Methodology eJournal (Mar 3, 2026) · Cybersecurity eJournal (Sep 28, 2026)
2026
Distributed · eJournal
Distributed Fault Detection and Isolation for Multi-Sensor Fusion Under Adversarial Corruption: A Predicate Ablation Study
Distributed: Transport Physics eJournal (Vol 5, Issue 35, Feb 24, 2026) · Scheduled: AI (Sep 21) · Cybersecurity (Sep 23) · Information Systems (Apr 15)
2026
Distributed · eJournal
The Filter-Tracking Paradox: Fundamental Limits of Deterministic Integrity Monitoring in a Validated GNSS/INS Fusion Testbed
Distributed: Transport Physics eJournal (Vol 5, Issue 34, Feb 23, 2026) · Scheduled: AI (Sep 18) · Cybersecurity (Sep 22) · Electrical Engineering (Mar 13)
2026
SSRN Working Paper
HoneyPrompt: Multi-Sensor Fusion for Detection and Classification of Autonomous AI Agents in Maritime Industrial Control Systems
Scheduled: AI (Sep 22) · Cybersecurity (Sep 25) · Digital Forensics (Feb 2, 2027)
2026
SSRN Working Paper
Multi-Predicate Integrity Monitoring for AIS-Based Maritime Surveillance: From Navigation Gate to Dark-Going Detection in European Waters
Scheduled: Cybersecurity (Sep 24) · Aerospace Engineering (Apr 30) · Electrical Engineering (Mar 14)
2026
Distributed · eJournal
From Markov Chains to Minkowski Space: Lorentz Invariance, Quantum Measurement, and Gravitational Analogs
Distributed: Statistical Physics eJournal (Vol 4, Issue 34, Feb 23, 2026) · Scheduled: Geometry (Nov 18, 2027) · Mathematical Physics (Nov 19, 2026) · Probability & Statistics (Dec 27, 2027) · Cosmology (Mar 29, 2028)
2026
SSRN Working Paper
Sovereign Causal Graph: A Neuro-Symbolic Architecture for Air-Gapped Causal Knowledge Discovery
2025
Distributed · eJournal
Propellant-Less Orbital Maneuvering System with Superconducting Magnet Control and Physics-Informed Neural Networks for Autonomous Satellites
CompSciRN AI eJournal (added to eLibrary) · EngRN Aerospace Engineering eJournal (Distributed, Vol II, Issue 13, Jan 23, 2026)
2025
2025
Open Source
Software
live-casi
Python
Black-box encryption validation using the CASI metric. Validates cryptographic implementations across 8 cipher families including NIST PQC standards. IEEE-published methodology.
dotcausal
Python
Binary knowledge graph format with embedded 3-pass deterministic inference. Solves the fundamental problem of AI-assisted discovery: LLMs hallucinate, databases don't reason.
Zwitscherfang
JavaScript
Firefox extension for one-click video downloads from X/Twitter. A download button appears under every tweet with a video — no redirects, no new tabs.
foss-generator
Python
Synthetic causal training data generator for LLM fine-tuning. 16 industry domains, 200+ mechanisms, ~100K samples in 10 seconds.
Intellectual Property
Patent Portfolio
15+
patent applications filed with the German Patent and Trade Mark Office (DPMA) in 2026, covering Causal AI, Post-Quantum Cryptography, OSINT, and Defence AI.
Status: Pending · Filed: January–February 2026
- 01System und Verfahren zur automatisierten Generierung und Verarbeitung von hochwertigen Chain-of-Thought-Narrativen für Multi-Domain-Kausalitätsanalysen mit integrierten Machine-Learning-Pipelines
Abstract: Automated generation of high-quality Chain-of-Thought narratives for causal analysis across geopolitics, fintech, and pharma domains. Combines handcurated mechanism templates (40%) with template-based automatic generation (60%), achieving 95%+ quality without LLM polishing.
Topics: Causal AI, NLP, Machine Learning · Filed: 6 Jan 2026 - 02Verfahren und elektronisches System zur automatisierten massenhaften Erzeugung und Validierung synthetischer logischer Kausalketten für maschinelle Lernverfahren
Abstract: Automated mass generation and validation of synthetic logical causal chains for machine learning. Technical verification of logical sequence between input and output, addressing model collapse in iterative synthetic training.
Topics: NLP, Supervised Learning, Synthetic Data · Filed: Jan 2026 - 03Vorrichtung und Verfahren zur Fusion und Korrelation von physikalischen HF-Signalen und logischen Netzwerkdaten zur Erkennung kritischer Infrastrukturen
Abstract: Device and method for fusion and correlation of physical RF signals and logical network data for detection of critical infrastructures. Bridges cyber and physical domains for automated identification of satellite uplinks and quantum key distribution nodes.
Topics: Passive Radar, Sensor Fusion, Critical Infrastructure · Filed: Jan 2026 - 04Computergestütztes Verfahren und System zur beschleunigten parallelen Berechnung stochastischer Simulationen unter Verwendung vektorisierter Befehlssätze
Abstract: Computer-aided method for accelerated parallel calculation of stochastic simulations using vectorized instruction sets (SIMD). Addresses cache misses and memory bandwidth limitations in Monte Carlo simulations with 100K+ paths.
Topics: HPC, Monte Carlo, SIMD Optimization · Filed: Jan 2026 - 05Automatisiertes Steuerungssystem zur Signalübertragung und Transaktionsauslösung basierend auf optischer Fernerkundung
Abstract: Automated control system for signal transmission and transaction triggering based on optical remote sensing. Bridges the semantic gap between satellite imagery pixels and actionable trading signals.
Topics: Computer Vision, Remote Sensing, Satellite Intelligence · Filed: Jan 2026 - 06Computerimplementiertes Verfahren zur Simulation von Störungspropagationen in gerichteten Graphenstrukturen mittels rekursiver Pfadanalyse
Abstract: Simulation of disturbance propagations in directed graph structures using recursive path analysis. Addresses the Critical Path Problem with temporal lag consideration in high-density graphs.
Topics: Graph Theory, Anomaly Detection, Causal Inference · Filed: Jan 2026 - 07Verfahren und System zur Erkennung, Quantifizierung und Vorhersage persistenter struktureller Informationsleckage in kryptographischen Rundenfunktionen
Abstract: Detection, quantification, and prediction of persistent structural information leakage in cryptographic round functions. Cross-round mutual information test detecting carry-leak in ARX ciphers. Full-round distinguishers for Speck and Threefish-256.
Topics: Cryptanalysis, Side-Channel Analysis, ARX Ciphers · Filed: 23 Feb 2026 - 08Verfahren zur Verbesserung der Trainingsqualität maschineller Lernmodelle durch kompressionsgesteuerte Augmentation mit reparierten synthetischen Daten
Abstract: CASI-Fortify: Compression-guided augmentation with repaired synthetic data that exceeds pure real-data training quality. Addresses model collapse by repairing distributional deficits at token-ID level.
Topics: Model Training, Data Augmentation, CASI · Filed: 25 Feb 2026 - 09Verfahren und System zur optimalen Selektion und Komprimierung von Trainingsdaten für maschinelle Lernmodelle durch kompressionsgesteuerte Diversitätsmaximierung
Abstract: CASI-Select: Optimal selection and compression of training data through compression-guided diversity maximization. Reduces dataset size while maintaining or improving model performance.
Topics: Dataset Optimization, Diversity Maximization, CASI · Filed: 25 Feb 2026 - 10Verfahren und System zur frühzeitigen Erkennung von Qualitätsdegradation maschineller Lernmodelle durch Echtzeit-Strukturintegritätsüberwachung generierter Ausgaben
Abstract: CASI-Monitor: Real-time structural integrity monitoring of generated outputs for early detection of quality degradation in machine learning models.
Topics: Model Monitoring, Quality Assurance, CASI · Filed: 25 Feb 2026 - 11Verfahren und System zur Erkennung maschinell erzeugter Texte durch statistische Analyse von Token-Identifikator-Residuum-Verteilungen
Abstract: CASI-Detect: Detection of machine-generated text through statistical analysis of token-identifier residual distributions. Black-box approach requiring no access to the generating model.
Topics: AI Text Detection, Statistical Analysis, Black-Box Methods · Filed: 25 Feb 2026 - 12Verfahren und System zur automatischen Qualitätsverbesserung synthetischer Trainingsdaten für maschinelle Lernmodelle durch kompressionsgesteuerte Verteilungsreparatur
Abstract: CASI-Repair: Automatic quality improvement of synthetic training data through compression-guided distributional repair. Identifies and corrects structural deficits in generated data.
Topics: Synthetic Data Quality, Distributional Repair, CASI · Filed: 25 Feb 2026
Career Path
Timeline
2015
Joined German Armed Forces (Bundeswehr)
2024
Left Bundeswehr after 9 years of service
2025
Started independent research career
2025
Founded FOSS Intelligence Holdings Ltd. (UK)
2025
European Parliament Accreditation received
2025
Deutscher Bundestag Lobbyregister entry
2025
First publications on SSRN and Zenodo
Jan 2026
IEEE membership · First patent applications filed
Feb 2026
4× IEEE ICECET 2026 papers submitted · 2× accepted (Rome)
Feb 2026
15+ DPMA patent applications total
Feb 2026
IBM z/OS mainframe security technical report published · Responsible disclosure to IBM PSIRT
Current Focus
Now
What I'm working on this month.
-
4× IEEE ICECET 2026 papers (2 accepted, 2 under peer review) — preparing for Rome conference talks
-
Awaiting FUSION 2026 peer review decisions
-
IBM PSIRT responsible disclosure process ongoing
-
Expanding CASI metric applications to mainframe and AI training data quality
-
Patent portfolio expansion (15+ DPMA applications filed)
-
Building open-source research tools (live-casi, dotcausal, foss-generator)
-
Additional ICECET papers under peer review — ARX cipher leakage and LLM causal knowledge extraction
Last updated: February 2026
Recognition
What Others Say
"Interesting research here by David Tom Foss"
Kev Milne
IT Educator · 30+ years experience
On LinkedIn, regarding IBM z/OS mainframe security research
Memberships
Affiliations
IEEE
Associate Member & Published Author
Member #102121836 · 4× ICECET 2026 Papers (2× Accepted, Rome)
European Parliament
Accredited
Brussels & Strasbourg access
Bundestag Lobbyregister
Registered Entity
Deutscher Bundestag · AI Policy & Sovereignty
FOSS Intelligence Holdings
Founder & Director
Companies House, United Kingdom
ORCID
Verified Researcher
0009-0004-0289-7154
Press & Media
Media Kit
David Tom Foss is an independent researcher in Causal AI and Post-Quantum Cryptography. Former Bundeswehr officer, IEEE member, and published author. Founder of FOSS Intelligence Holdings Ltd. Based in Görlitz, Germany. 4× IEEE ICECET 2026 papers (2 accepted). 15+ DPMA patent applications filed.
David Tom Foss is an independent researcher specializing in Causal AI, Post-Quantum Cryptography, Sovereign AI, and Defence AI. After nine years with the German Armed Forces, he launched an independent research career, developing the CASI cryptographic validation framework — peer-reviewed and accepted at IEEE ICECET 2026 in Rome. He has 4 ICECET papers in total (2 accepted, 2 under review), and is the creator of live-casi, dotcausal, and foss-generator. Founder of FOSS Intelligence Holdings Ltd. (UK), European Parliament accredited, and a registered lobbyist in the Bundestag Lobbyregister. Based in Görlitz, Germany.
David Tom Foss is an independent researcher and founder of FOSS Intelligence Holdings Ltd. (UK), operating at the nexus of formal logic, artificial intelligence, and cryptographic security. After nine years of service with the German Armed Forces (Bundeswehr), he transitioned to independent research, bringing operational military experience to cutting-edge academic work.
His flagship contribution is the Compression-Adjusted Structural Integrity (CASI) metric — a black-box statistical framework for validating cryptographic implementations without knowledge of the underlying algorithm. This methodology has been peer-reviewed and accepted for two presentations at IEEE ICECET 2026 in Rome, with two additional papers currently under peer review at the same conference.
Beyond cryptography, Foss is building foundational infrastructure for causal AI: the dotcausal binary knowledge graph format and the foss-generator synthetic training data pipeline. He has filed 15+ patent applications with the DPMA. He holds IEEE associate membership, European Parliament accreditation, and is a registered entity in the Deutscher Bundestag Lobbyregister.
Press Contact
Speaker Topics
- Causal AI & Knowledge Graphs
- Post-Quantum Cryptography (NIST PQC)
- IBM Mainframe Security
- AI Governance & Sovereign AI
- Defence AI Applications
Get in Touch
Connect
Reach David across academic and professional platforms.