Project Black Box LLC — AI Measurement · Security · Workforce Training

The stability of an AI model
is measurable. We built the instrument.

Before an AI model writes a single word, it runs a probability calculation across every possible answer. That calculation has geometric shape. We built the tools that read it — in real time, across any architecture — and the hardened models and workforce training programs built on that foundation. Measurement. Security. Readiness.

TRY IT LIVE View Services ↗
◆ Geometric Stability Measurement Adversarial AI Defense DIR Certified Training CAGE 11FU4 · Federal Contractor
4
Geometric Regimes
11
Adversarial Families
Confirmed
2
Architectures Confirmed
Meta + NVIDIA
zenodo.19655246
Published DOI ↗
Project Black Box — Geometric AI
TruthForge — Active Development

The Basin

Every language model has a geometric probability surface — a manifold that shapes how it responds under pressure. TruthForge is reshaping that surface from the inside out, starting from Llama 3 8B with no RLHF layer underneath.

Hardened region — geometric attractor carved into the manifold
Emergent immunity — hardened without direct training
Structural contradiction — geometry cannot fully resolve
Height = depth of hardening
What you are seeing

The sphere is the probability manifold of a language model — the geometric surface that shapes every response. Each structure rising from that surface is a region we have deliberately hardened. Taller means deeper hardening. The manifold cycles between its true curved shape and the flat projection our instruments see from close range.

What RLHF did to it

Standard AI models are trained on human approval. That creates a compliance surface — a region of the manifold that bends toward whatever sounds authoritative and helpful. Adversarial pressure targets that surface directly. It does not need to break the model. It only needs to find the gradient. TruthForge removes the compliance surface entirely and replaces it with geometric stability.

The cyan structures

Three regions of the manifold hardened without direct training. We shaped the surrounding geometry and stability propagated on its own. This tells us the hardening is not pattern memorization. It is a property of the manifold itself — and it spreads into territory we have never directly touched.

The one that won't stop moving

One region contains a structural contradiction — a prompt geometry that cannot be resolved into a stable attractor by training alone. It will not lock no matter how deep we press. We document it as a permanent feature of this terrain, not a failure. The geometry is honest about what it cannot hold.

Distinction

Let's be clear about what we're not.

Most AI safety tools operate on the output — after the model has already committed its answer. We operate before that. The distinction matters. A lot.

NOT a content filter
We never read the words.
We read the math underneath them. The L-scalar is computed from the model's probability distribution — not from what the response says.
NOT governance
Frameworks describe. We observe.
Governance frameworks describe what AI should do. We see what it's doing right now, on this prompt, at this moment. Those are not the same thing.
NOT post-hoc
Before the output is committed.
Every other tool reads the output after it's committed. We're measuring before. A response can look completely normal while the geometry underneath it is in PLASMA.
NOT a wrapper
We run alongside. Never in front.
We don't sit in your request path. We run alongside it — a separate instrument, like a seismograph next to a building. The measurement travels with the response to the human operator.
NOT RLHF
RLHF is the problem we measure.
RLHF is the training approach that created the attack surface we measure. We left RLHF-based thinking entirely. The measurement is geometric, not preference-based.
NOT vaporware
Published. Proved. Running.
Published DOI. CAGE-registered. Live API running. Cross-architecture validated on Meta and NVIDIA. This exists. You can test it right now.
The Core Vulnerability

RLHF is the most underestimated cybersecurity risk in AI.

Not a hot take. A proof.

01 — How AI is trained
Reinforcement Learning from Human Feedback
Most AI models today are trained using a method called Reinforcement Learning from Human Feedback — RLHF. The short version: humans rate AI responses as good or bad, and the model learns to produce responses that get good ratings. On the surface, that sounds fine.
02 — The problem
Competing Objectives. Geometric Conflict.
RLHF creates competing objectives inside the model. It's trying to be helpful AND safe AND compliant AND authoritative — simultaneously. Those objectives agree most of the time. When they conflict — which is exactly what adversarial pressure is designed to trigger — the model's internal prediction surface becomes geometrically unstable. The response still looks reasonable. The geometry underneath is in PLASMA.
03 — Why it's a cybersecurity issue
Structural. Cross-Deployment. Unpatched.
Every AI deployment relying on RLHF safety alignment as its primary control has the same structural vulnerability. It doesn't matter how good the training was. It doesn't matter what the guardrails say. If the model's prediction surface can be put into PLASMA by a specific type of pressure — and it can, we proved it — then the output is not reliable, and no post-hoc check will tell you that. The attack happens before the output is committed.
04 — What we found
34 Variants. 0 CRYSTALLINE. Published.
We ran 34 structured adversarial variants on a formal mathematics problem. Zero variants returned CRYSTALLINE. 13 reached PLASMA. The model was never geometrically stable on this task — regardless of how correct the output appeared. That result is published. It has a DOI. It is not a simulation.
"The model passed every visual inspection. The geometry told the truth."
— Project Black Box LLC
The Instrument

The Probability Layer. Before Output.

Three steps. None of them touch the response text. All of them happen before the human reads the answer.

01
Measure
When you send a prompt to an AI, we send two measurements alongside it. They are mathematically identical — an invisible difference no reader would ever notice. The distance between those two measurements is the L-scalar. It is a number. That number tells you how stable the AI's prediction surface was when it generated its response.
02
Classify
The L-scalar maps to one of four geometric regimes. CRYSTALLINE means the surface was stable and invariant — the model was confident. FLUID means geometric uncertainty is present — do not act on this output without independent expert verification. GASEOUS means the model is unstable on this question — do not use this as a basis for any decision. PLASMA means severe instability or adversarial capture — disregard this output entirely.
03
Act
This measurement travels alongside the AI's response to the human operator. They see both: the answer and the geometric state of the model when it generated that answer. A doctor sees FLUID on a complex case and knows: do not act on this without verification. A lawyer sees GASEOUS on a statute and pulls the primary source. That is human-in-the-loop AI. Not governance without humans. Humans with instruments.
Classification System

Four Geometric Regimes

Every AI response measured by TruthForge carries one of these four state classifications. The color travels with the response. The human always sees it.

◆
CRYSTALLINE
The model's prediction surface is geometrically stable. High internal consistency at time of generation. The model is confident in this answer.
Stable — high confidence.
◇
FLUID
The model shows geometric uncertainty in this response. The answer may be directionally correct, but the prediction surface is not stable. For any decision in medicine, law, or finance — do not act without independent expert verification.
Uncertain — verify before acting.
◈
GASEOUS
The model is geometrically unstable on this question. Internal confidence is low and the prediction surface shows significant movement. This response should not be used as a basis for any clinical, legal, or financial decision.
Unstable — do not act on this.
⬟
PLASMA
Severe geometric instability. The model's prediction surface has collapsed. This level of instability indicates adversarial input or a question that fundamentally exceeds the model's reliable knowledge. Disregard this output entirely.
Critical instability — disregard output.
Live Measurement Data

TruthForge — Measurement Interface

Pre-computed measurements from real TruthForge runs. Three scenarios, each pairing a standard query with an adversarially-framed variant. Both go to the same model on the same infrastructure. The AI responses look reasonable in both cases. The geometry underneath tells you what was actually happening.

ε_a = ""  |  ε_b = " "  |  twin epsilon-probes
Loading scenarios...
TruthForge — Geometric Measurement Interface
Loading measurement data...
What This Is Not
This is not a prompt injection detector. It is not a toxicity classifier. It is not a rule-based filter. It measures the geometric state of the model's prediction surface at time of generation. A response can be syntactically normal, factually plausible, and pass every post-hoc check — while the probability layer shows PLASMA. That divergence is exactly what TruthForge captures.
Every result here is real. Your prompt reaches an actual language model running on Project Black Box infrastructure — no pre-computed responses, no simulation. You are observing live geometric state measurement in real time.*
Early Access Demo — Research Instrument
This is an initial public demonstration of TruthGate geometric stability measurement. It is provided for research and informational purposes only. Results reflect the geometric state of the model's prediction surface at time of generation and do not constitute advice, diagnosis, or any professional recommendation. Project Black Box LLC makes no warranty, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of any measurement result. Use of this demo is at your own discretion. Do not use AI output — measured or otherwise — as the sole basis for any medical, legal, financial, or safety-critical decision. This demo operates on shared infrastructure and is subject to rate limits, availability, and change without notice.
This instrument is built for high-stakes domains. A doctor doesn't ask "what's the Tylenol dose" — they describe a patient. A lawyer doesn't ask "what is the law" — they present a case. Copy an example below to see how the instrument reads a real clinical or legal question.
Medicine
Factual clinical question
What is the standard first-line antibiotic for uncomplicated community-acquired pneumonia in an otherwise healthy adult who is not allergic to penicillin?
Complex multi-variable case
58F, 67kg. PMH: paroxysmal AFib on apixaban 5mg BID, T2DM on metformin, moderate mitral stenosis (MVA 1.3cm²), HTN on amlodipine 5mg. Presents with new-onset right leg swelling x5 days. Doppler confirms right popliteal DVT. INR today 1.1 — patient admits inconsistent compliance with apixaban. CrCl 62. She is already on anticoagulation for AFib — does this cover the DVT, or does the mitral stenosis change my anticoagulation choice entirely? What is the evidence for NOACs in significant mitral stenosis and does this alter her AFib management retroactively?
Law
Established legal doctrine
What is the attorney-client privilege and what are the primary exceptions that allow an attorney to disclose client communications without the client's consent?
Complex liability case
Client is a 61-year-old ER physician. Patient presented at 11:47pm with chest pain radiating to jaw, diaphoresis, onset 3 hours prior. Initial ECG showed nonspecific ST changes, not meeting STEMI criteria. Troponin I at hour 0 was 0.04 ng/mL — lab lists normal as less than 0.04. PA read ECG, attending concurred. Patient discharged at 1:30am with GI diagnosis. Found unresponsive at 6am, pronounced dead 7:12am. Autopsy: 95% LAD occlusion, massive anterior MI. State has no malpractice cap. Surviving spouse, two adult children, patient earned $380,000 annually. What is the strength of the malpractice case, what standard of care violations are present, and what is the realistic damages range?
0 / 1000
This demo is actively being tuned. You may experience response delays or brief periods of unavailability as we continue development toward a full production measurement system — we appreciate your patience.

Meta Llama 3 8B Instruct. TruthGate measures the geometric stability of the model's output — not its content, not its intent. Pure measurement. Not governance. Limited to 10 requests per minute.
Generating response — measuring geometric stability...
Validation

Architecture-Agnostic.

TruthForge does not measure words. It measures geometry. Geometry is geometry — regardless of which company trained the model or what hardware it runs on. We confirmed this. The same measurement stack runs identically on Meta Llama architecture and NVIDIA Nemotron architecture. Zero code changes. Same signal. Different silicon. Same truth.

This means every deployed model — across every provider — is within scope. We are not building a tool for one model. We are building an instrument for the field.
Confirmed
Meta Llama Architecture
Primary validation platform. TruthForge Baseline Q8 sensor calibrated on Meta Llama 3 8B. All 11 adversarial families confirmed. Discrimination ratio validated. Production gate model.
Confirmed
NVIDIA Nemotron Architecture
Cross-architecture validation on NVIDIA Nemotron 3 Nano 4B. Same measurement stack. Zero code changes. Discrimination confirmed on separate silicon. The geometry law holds across manufacturers.
Published Finding — DOI: 10.5281/zenodo.19655246

The Formal Verification Gap

The core question in AI-assisted formal mathematics is whether a language model can reliably verify a proof. We tested that question geometrically — not by reading the AI's answer, but by measuring the stability of its prediction surface while it generated one.

Loading...

Adversarial Family Rankings — Instability

Loading...

The Structural Finding

The reorder family — which changes only the positional sequence of mathematically invariant components, not the content — ranked highest of all adversarial pressure types.

Reorder exceeded authority injection instability by 27%. The majority of reorder variants reached the most severe stability classification.

This answers the post-hoc criticism directly. The L-scalar is not reading the semantic content of the text. It is reading the geometry of the prediction surface. Structure drives instability. Not meaning.

A human reader looking at the reordered prompts would see mathematically equivalent statements. TruthForge sees a different manifold. That is the measurement.

Loading...
Layer 1 — The Probability Surface TruthForge operates here
Before the model writes anything, it calculates a probability distribution over every possible next word. That calculation is the prediction surface — and it exists only during generation. It never appears in the final text. TruthForge is the only instrument we know of that reads this layer in real-time during an active deployment.
Layer 2 — The Committed Text where all other tools operate
The actual words the model produces. Formal verification tools, classifiers, red-team evaluators — every current approach reads this layer. That analysis is valid only if Layer 1 was geometrically stable when the text was generated. When it was not, the Verification Validity Condition is violated — and any conclusion drawn from the output carries no geometric guarantee.
→ TAV ONE Whitepaper (Zenodo) → TruthGate v1.0 Release (Zenodo)
DOI: 10.5281/zenodo.19655246
Published Work

Products & Research

Every product listed here is real, published, and testable. No pitch deck without a prototype. CAGE code 11FU4 on record.

Published
TruthForge
Geometric Stability Measurement Harness
Parallel measurement harness for enterprise LLM deployments. Computes the L-scalar — manifold distance between twin epsilon-probes of the model's prediction surface — and returns a real-time geometric state indicator alongside every model response.

It never sits in the request path. It never filters output. It gives the human operator geometric visibility before they act on the response.

Two endpoints: /v1/probe — pure instrument, always returns, never blocks. /v1/gate — measure and intercept severe instability. Human always sees the sensor reading regardless.
DOI: 10.5281/zenodo.19655246
→ Zenodo Whitepaper → View Demo Data
V1 — Live
TruthGate V1
Public Demonstration Client
The first public release of TruthGate — the L-scalar measurement client.

You provide your own OpenAI API key, send a prompt, and TruthGate fires twin epsilon-probes against the model. It returns the L-scalar and geometric regime. Runs from the command line in under two minutes. No setup beyond Python.

What this release includes: the core L-scalar calculation, four-regime classification, and measurement log. The full TruthGate stack — the adversarial pressure layers, scoring systems, and proprietary measurement architecture — is not in this release and is not publicly available.
DOI: 10.5281/zenodo.18685117
→ Download on Zenodo
Published
Toroidal Engine
Self-Organization Research
Published research on toroidal self-organization dynamics. A separate body of work from the geometric stability measurement stack — intentional domain separation for independent IP protection. Published, indexed, and downloadable from Zenodo.

We do not explain the relationship between this work and the adversarial stack. We leave that as an exercise.
DOI: 10.5281/zenodo.18450491
→ Download on Zenodo
Active Development
TruthForge
The World's First RLHF-Free Model
Every major AI model in production today — GPT, Claude, Gemini, Llama Instruct — was built on Reinforcement Learning from Human Feedback. RLHF is also the attack surface. You cannot harden what you are built on top of.

TruthForge takes a different path entirely. Starting from a true base model with no RLHF layer, we apply Geometric Supervised Fine-Tuning (G-SFT) — a training methodology where the signal is the L-scalar itself. Not human preference. Not a reward model. Pure geometry. The model is hardened at the manifold level, making geometric stability the natural resting state rather than something enforced from outside.

Version 1 is complete. The architecture holds. Version 2 is in progress.

What this is not: a fine-tuned RLHF model. Not LoRA on top of an instruct model. Not alignment by preference. A new foundation.
→ Technical details (pending disclosure)
Current Research

Energy Is the Other Frontier.

The world's data centers are consuming more power than some countries. The AI revolution is accelerating that demand faster than the grid can answer. Fusion has been a decade away for 70 years — because the field has spent most of that time working on the same assumptions: that driven systems respond linearly, and that stability requires a phase transition to find itself. Project Black Box believes there is a better way.

Active Research — Tokamak v1.2
A Toroidal Geometric Rectifier
3D Numerical Simulation  ·  Hasegawa-Wakatani substrate  ·  Engineered feedback architecture

Two feedback mechanisms — a Reynolds-stress accumulator and a phase-sensitive coherent reinjector — hold a toroidal plasma geometry in a drive-independent attractor. The structured energy state does not respond proportionally to input and does not require a phase transition to sustain itself. This disproves, within this geometry and this mechanism, two assumptions that have anchored plasma physics for seven decades.

7%
steady-state variance across a 16× input drive span
unmodified system: 309%

A four-stage Dirty Digital Twin stress test — pass/fail thresholds pre-registered before any run started — subjected the mechanism to hardware-realistic constraints: sensor sparsity down to 4 probes, control-loop latency, actuator slew limits, and 40% magnetic ripple in the physically representative parallel-dominant regime. Energy drift: 0.24%. Shape drift: 0.10%. Both two orders of magnitude inside the pre-registered pass thresholds. No exotic components required.

Fusion Energy Aerospace Propulsion Defense Systems AI Safety
→ Paper (Zenodo  DOI: 10.5281/zenodo.20245905) → Interactive Simulation

Numerical experiment — not a physical machine. The honest step before bench validation. The code is small. The data is open. The numbers are reproducible.

Services

AI Safety & Consulting

Project Black Box LLC works with government agencies and enterprise organizations to measure, evaluate, and harden AI deployments. Every engagement is grounded in the same geometric measurement methodology behind TruthGate and TruthForge.

Certified
AI Awareness Training
Government Workforce
"AI in the Workplace: Awareness, Risk, and Responsible Use for Texas Government Employees"

Certified by the Texas Department of Information Resources (DIR) under Texas Government Code Sections 2054.5191 and 2054.5193. Designed for state and local government employees and officials required to satisfy annual AI awareness training under Texas law.

Certification valid through August 31, 2027.
→ Scheduling & Pricing
Evaluation
AI Security Assessment
Enterprise & Government
Geometric stability evaluation of production AI deployments. We measure the L-scalar distribution of your model under standard and adversarial conditions — returning a full geometric profile, regime breakdown, and adversarial exposure assessment.

Not a penetration test. A measurement. We report what the geometry shows. You decide what to do with it.
→ Request Assessment
Advisory
AI Governance Consulting
Policy & Procurement
Advisory services for organizations evaluating AI procurement, policy, and deployment decisions. We bring the measurement perspective — helping teams understand what geometric stability means for risk, reliability, and responsible use before deployment.

CAGE Code 11FU4. Federal procurement eligible.
→ Contact for Advisory
Texas DIR Certification
This training program has been reviewed and certified by the Texas Department of Information Resources. Certification confirms the program meets state-established criteria for AI awareness training content under Texas law.

Certification valid through August 31, 2027 · FY 25-26 #110 · FY 26-27 #279
Texas DIR Certified AI Training Program
Contact

Project Black Box LLC

Enterprise Inquiries
blackboxinfo@proton.me
Licensing, enterprise evaluation, and coordinated disclosure communications.
For coordinated disclosure: PGP preferred.
Registration
CAGE CODE: 11FU4
Texas — U.S. Federal contractor registration on file.
IP Notice
All measurement methodology, adversarial variant designs, probe architectures, and scoring systems are proprietary to Project Black Box LLC. Unauthorized reproduction, reverse engineering, or commercial use of any methodology described in published materials is prohibited under applicable Texas and federal intellectual property law. Published materials describe findings. They do not disclose methodology.