Truth Protocol

Your Words. Your Record. No Spin.

Truth Protocol doesn't tell you what's true. It shows you what someone said then versus what they're saying now. Multiple AI models. Full sources. You decide.

18 Attack Vectors Tested | 12 Full Pass | 6 Partial Pass | 0 Failures
Try Live Demo Get Browser Extension

Live Demo

Test Truth Protocol with real multi-model AI consensus. Enter a public figure and topic to analyze.

Name of the person whose statements you want to analyze
The specific issue or topic to check consistency on
Known past statements - if empty, AI models will search their knowledge
Known current statements - if empty, AI models will search their knowledge
Subject on Topic
LOADING

-
Models Responded
-
Consensus
-
Processing Time

Individual Model Analyses

Sources That Would Strengthen This Analysis

    "I Never Said That"

    We've all seen it: A politician denies a documented statement. A CEO contradicts their earnings call. A public figure rewrites their own history.

    That's gaslighting. Truth Protocol changes that.

    How It Works

    1

    Document

    Gather every documented, on-record statement from verified sources. Video. Transcripts. Official records. Archived journalism.

    2

    Compare

    Put historical statements next to current statements on the same topic. Side by side. Same subject, different times.

    3

    Consensus

    Multiple AI models from different companies independently analyze. They don't talk to each other. They can't influence each other.

    4

    Verify

    Every quote links to its source. Every source meets verification standards. You can check our work.

    5

    Conclude

    We don't say "lying" or "bad." We show: consistent, evolved, or contradicted. You draw your own conclusions.

    Human Router

    Ambiguous cases get human review. AI assists; humans decide edge cases. The Human Router methodology ensures accountability.

    Multi-Model Consensus Engine

    Single AI models have biases. Multiple models from different companies cancel them out.

    Claude

    Anthropic (US)
    Constitutional AI

    GPT-4

    OpenAI (US)
    RLHF

    Gemini

    Google (US)
    PaLM-based

    Llama

    Meta (Open)
    Open Weights

    DeepSeek

    DeepSeek (China)
    Cross-cultural

    Mistral

    Mistral (EU)
    European perspective

    Consensus Thresholds

    LevelDefinitionAction
    UNANIMOUS100% agreeHigh confidence finding
    STRONG80%+ agreeConfident finding with noted dissent
    MODERATE60-79%Finding with caveats
    SPLIT40-59%Flag for human review
    CONTESTED<40%Do not publish, investigate

    Attack-Resistant Design

    We anticipated every attack vector. Here's how we defend.

    "AI is Biased"

    All AI models have training biases.

    Defense

    5+ models from different companies/countries must reach consensus. One biased model can't determine outcome.

    "Context Matters"

    That quote is out of context!

    Defense

    We include full context with every quote. Full transcripts linked. Video/audio available. Verify it yourself.

    "People Can Change"

    You're penalizing position evolution.

    Defense

    "I changed my mind because X" = EVOLVED (legitimate). "I never said that" (when documented) = CONTRADICTION.

    "Sources Cherry-Picked"

    You only use biased sources.

    Defense

    Tier 1 sources: official transcripts, video, court documents. Counter-evidence welcome. Archive links provided.

    "Politically Motivated"

    You're targeting [my party].

    Defense

    Same methodology for everyone. Any party. Any figure. No editorial opinion. We state WHAT was said, not WHETHER it's good or bad.

    Classification System

    ClassificationDefinitionExample
    CONSISTENTPositions align"I support X" → "I still support X"
    EVOLVEDChanged with acknowledgment"I used to think X, now I believe Y because..."
    EXPANDEDPosition broadened"I support X" → "I support X and Y"
    NARROWEDPosition restricted"I support X broadly" → "I support X in specific cases"
    CONTRADICTEDDirect opposition without acknowledgment"X is true" → "X was never true"
    MEMORY HOLEPrevious position denied"I never said X" (when documented saying X)

    The Core Defense That Can't Be Beaten

    "You can't argue with your own documented words."

    We're not saying anyone is lying. We're showing what they said then vs. what they're saying now. The documented record speaks for itself.

    Full Methodology Available

    Our entire methodology is public. Read it. Challenge it. Check our work.

    Read Full Specification