Independent AI Systems Auditor · AI Red Teaming · Deterministic Security Architecture
Poland · English / Polish / German
GitHub: @cryptofixyup · Email: kpiotr210@gmail.com
I operate as an independent technical assurance practitioner examining whether AI-enabled systems are correctly specified, securely bounded, reliably implemented, and backed by reproducible evidence. My work combines adversarial model-behavior testing with systems programming (Rust, Python, TypeScript), zero-trust architecture, and strict evidentiary audit standards (E0–E5).
I do not claim a PhD, university degree or formal institutional affiliations. My methodology and tooling have been developed through independent practitioner research, hands-on red teaming, and open technical verification.
My work enforces an explicit six-level evidentiary hierarchy:
- E0 (Unsubstantiated): Unverified assertions, promotional claims, or marketing metrics.
- E1 (Assertion): Stakeholder descriptions without independent reproduction data.
- E2 (Documentary): S-1 filings, architectural specifications, design documents, and static snapshots.
- E3 (Artifact): Inspected source code, schema definitions, build scripts, and structured test configs.
- E4 (Reproduced): Independent test executions reproduced in isolated, controlled environments.
- E5 (Operational): Continuous telemetry, immutable ledgers, and verified production runtimes.
Core Principle: "Systems don't run on intent. Demand the receipt."
- AI Systems & Agent Auditing: Independent verification of agentic workflows, deterministic action gates, and tool permission boundaries.
- Adversarial Red Teaming: Multi-turn jailbreak dynamics, refusal-mechanism boundary analysis, prompt-injection defenses, and context-retention failure modes.
- Deterministic Action Gates: Enforcing human-in-the-loop and cryptographic verification before any agent executes irreversible operations.
- Cross-Chain Signal Intelligence: Verification of multi-chain threat indicators (Time Sync, Funding Link, Bytecode Reuse, LP Pattern analysis).
- Supply-Chain & Runtime Hardening: Clean-room container builds, dependency pinning, cgroup resource bounding, and SIEM forwarding verification.
| Project / Repository | Focus & Verified Scope | Evidence Level |
|---|---|---|
| AI Systems Auditor Charter | Role specification, operating standard, finding taxonomy, and audit ledger protocols | E2 / E3 |
| Alpha Sentinel | Deterministic authorization controls and execution gates for autonomous agent systems | E3 |
| pelican-rs | High-performance Rust telemetry decoding with length, version, checksum, and semantic validation | E3 / E4 |
| market-monitor-rs | Rust market-signal parsing, fixed-depth order books, and deterministic sequence validation | E3 |
| Policy Activation Kernel | Dependency-free reference policy kernel and deterministic state transition vectors | E3 / E4 |
| OBLITERATUS | Public research workspace on LLM safety boundaries, refusal triggers, and adversarial prompt dynamics | E3 |
Note: Proprietary client audits and red-teaming evaluations remain in private repositories and are shared only under mutual confidentiality agreements.
- Evidence Before Assertion: A claim is treated as unverified until supported by a test log, bytecode match, or reproducible ledger.
- Deterministic Control Over Heuristics: AI may propose actions; deterministic policy engines decide and execute.
- Fail-Closed by Default: Any parsing mismatch, ambiguous instruction, or missing parameter immediately aborts the execution path.
- Separation of Concerns: Governance, authorization, execution, and verification operate as distinct structural layers.
- No Synthetic Pass Marks: Passing a static benchmark proves only that specific test vector; real-world safety demands adversarial pressure testing.
- Systems & Languages: Rust (1.75+), Python (3.10–3.12), TypeScript / JavaScript, Bash / POSIX Shell.
- Security & Networking: Raw socket parsers, async TCP scanners, deterministic rate monitors, EIP-712 signature verification, zero-trust envelopes.
- Infra & DevOps: Docker (cgroup v1/v2), Linux, GitHub Actions CI/CD, PostgreSQL, Redis, Kubernetes topologies.
- Operations Experience: ~25–27 years in logistics, warehouse operations, and transport systems (TSL), informing real-world operational failure analysis.
- Credentials: Independent software practitioner and red-team researcher; no institutional research funding.
For independent AI systems auditing, agentic red teaming, or deterministic security architecture reviews:
Email: kpiotr210@gmail.com
GitHub: @cryptofixyup