Medical and Data Disclaimer

1. Not Medical Advice

1.1 Intended Use

withZeta is designed for:

• ✓ Preclinical research exploration
• ✓ Drug discovery hypothesis generation
• ✓ Literature synthesis and evidence gathering
• ✓ Clinical trial identification (research purposes)
• ✓ Educational purposes and academic research

withZeta is NOT designed for:

• ✗ Diagnosing diseases in patients
• ✗ Prescribing treatments or medications
• ✗ Replacing clinical judgment or professional medical advice
• ✗ Direct patient care decisions
• ✗ Medical record management or clinical documentation

1.2 Healthcare Professional Requirement

DO NOT use Zeta outputs as a substitute for professional medical advice, diagnosis, or treatment.Always seek the advice of qualified healthcare providers with questions regarding medical conditions.

1.3 Emergency Medical Situations

If you or someone else is experiencing a medical emergency, call emergency services immediately (911 in the United States). Do NOT rely on withZeta for emergency medical guidance.

See also: Acceptable Use Policy

2. AI-Generated Content Limitations

2.1 Large Language Model Architecture

withZeta uses multiple AI models (both open-source and commercial) which may change as we improve our Services. These models have the following characteristics:

• Model Providers: We use various AI providers and may switch models without notice
• Context Window: Typically 200,000 tokens (approximately 150,000 words) but varies by model
• Training Data Cutoff: AI models have knowledge cutoff dates (may not include recent discoveries)
• Probabilistic Output: Responses are generated probabilistically, not retrieved from a database
• No Memory Between Sessions: Unless you explicitly continue a conversation, AI models do not remember previous interactions

2.2 Known Failure Modes & Hallucinations

Large language models can generate factually incorrect information ("hallucinations"). AI-generated responses may:

• Hallucination: Generate factually incorrect information presented with confidence
• Confabulation: Fill in missing details with plausible but fabricated information
• Invent Studies: Cite non-existent research papers or clinical trials
• Misinterpret Data: Draw incorrect conclusions from correct source material
• Outdated Information: Present historical data as current
• Logical Errors: Make reasoning mistakes despite having accurate facts
• Contradictions: Provide conflicting information within the same response
• Recency Bias: Overweight recent information in training data
• Prompt Sensitivity: Different phrasings of the same question may yield different answers
• Sycophancy: Tendency to agree with user assumptions rather than correct them

2.3 What AI Models Cannot Do

Despite advanced capabilities, AI models used in Zeta cannot:

• Access Real-Time Information: Cannot browse the internet or access data after training cutoff (except via tools we provide like PubMed API)
• Execute Code or Experiments: Cannot run clinical trials, validate findings experimentally, or synthesize compounds
• Guarantee Factual Accuracy: May generate plausible-sounding but incorrect information
• Understand Context Perfectly: May misinterpret ambiguous queries or nuanced medical context
• Access Proprietary Data: Cannot access unpublished research, proprietary drug data, or confidential trial results
• Replace Human Expertise: Cannot replicate clinical judgment, intuition, or patient-specific assessment

2.4 Not a Database Query System

Zeta is NOT a traditional database query interface. It:

• Interprets queries using natural language understanding (subject to misunderstanding)
• Synthesizes information from multiple sources (subject to integration errors)
• Generates prose explanations (not raw data dumps)
• May reformulate or simplify complex medical data (subject to oversimplification)

For precise data queries, consult original databases directly: PubMed, ClinicalTrials.gov, ORPHANET, etc.

2.6 Context Window Limitations

AI models used in Zeta have finite context windows (varies by model, typically around 200,000 tokens ≈ 150,000 words). This means:

• Very complex queries may exceed processing capacity
• Long conversation history may be truncated
• Knowledge graphs may simplify relationships to fit context
• Some nuances from source material may be lost in synthesis

The system will warn you when approaching context limits, but be aware that quality may degrade near these limits.

2.7 No Real-Time Data

While Zeta accesses external databases, there may be delays:

• PubMed abstracts are current, but full-text articles may lag publication dates
• Clinical trials database updated quarterly (not real-time)
• FDA drug labels updated when FDA publishes changes (not instantaneous)
• Rare cancers knowledge base updated monthly

3. Data Accuracy & Currency

3.1 Third-Party Data Sources

Zeta integrates data from external sources. We do NOT control these databases. Accuracy depends on source data quality:

PubMed/MEDLINE (U.S. National Library of Medicine)

• Coverage: 36+ million citations from biomedical literature
• Update Frequency: Real-time API access (current abstracts)
• Limitations: Abstracts only (full text requires subscription), not all journals indexed, pre-prints may not be included

ORPHANET (Rare Disease Database)

• Coverage: 9,000+ rare diseases
• Update Frequency: Continuous updates by ORPHANET team
• Limitations: European focus, may not include all ultra-rare conditions, data quality varies by disease

NCI Thesaurus (National Cancer Institute)

• Coverage: Cancer terminology and classification
• Update Frequency: Monthly releases
• Limitations: Terminology-focused (not treatment guidelines)

AACT Clinical Trials (ClinicalTrials.gov Snapshot)

• Coverage: 400,000+ clinical trials (global)
• Update Frequency: Quarterly snapshots
• Limitations: Data may be 1-3 months old, trials may be completed/suspended, contact info may be outdated

Rare Cancers Knowledge Base (Lantern Pharma Proprietary)

• Coverage: 438 rare cancer types with detailed molecular profiles
• Update Frequency: Monthly curation from literature
• Limitations: Curated data subject to interpretation, not comprehensive for all rare cancers, focused on therapeutic-relevant information

3.2 Data Staleness Warnings

Always check primary sources for time-sensitive information:

• Active Clinical Trials: Verify recruitment status directly with trial sites (trials can close rapidly)
• Recent Drug Approvals: Check FDA.gov for latest approvals (Zeta data may lag weeks)
• Treatment Guidelines: Consult NCCN, ESMO, or ASCO for current standard of care (guidelines update annually)
• Safety Alerts: Monitor FDA safety communications for drug recalls or black box warnings

3.3 Proprietary Data Limitations

Lantern Pharma's Rare Cancers Knowledge Base:

• Is curated from published literature (not exhaustive)
• Reflects therapeutic-focused data (may exclude non-therapeutic biomarkers)
• Covers 438 cancer types (not all ultra-rare subtypes included)
• Subject to curator interpretation and classification decisions

Not all rare cancers, biomarkers, or treatments are included in our database.

4. Blood-Brain Barrier Prediction

4.1 PredictBBB.ai Integration

Zeta integrates with PredictBBB.ai (external API) for blood-brain barrier permeability prediction:

• Accuracy: 94.1% on test dataset (NOT 100%)
• Method: Machine learning ensemble (16 models) trained on experimental data
• Input: SMILES string (molecular structure representation)
• Output: Binary prediction (BBB+ or BBB-) with probability score

4.2 Prediction Limitations

• Model trained on specific dataset (may not generalize to all drug classes)
• Does NOT account for active transport mechanisms
• Does NOT account for efflux pumps (P-glycoprotein, BCRP)
• Does NOT consider patient-specific factors (BBB disruption in tumors, age-related changes)
• Does NOT predict CNS distribution or brain tissue concentration

Use BBB predictions as screening tools, not definitive answers. Validate promising candidates with experimental assays.

4.3 Cheminformatics Limitations

Molecular descriptor calculations and SMILES validation:

• SMILES Validation: Checks syntax, NOT chemical feasibility or synthesizability
• Molecular Descriptors: Theoretical calculations (may differ from experimental values)
• Drug-Likeness Rules: Guidelines, not hard requirements (many successful drugs violate Lipinski's Rule of Five)

Real-world pharmacokinetics may differ significantly from in silico predictions.

5. Clinical Trials Information

5.1 AACT Database Limitations

Clinical trials data from AACT (Aggregate Analysis of ClinicalTrials.gov):

• Data is a quarterly snapshot (1-3 months old)
• Trials may be outdated, suspended, terminated, or completed since last snapshot
• Contact information may be stale
• Inclusion/exclusion criteria are summarized (not exhaustive)
• Recruitment status may have changed

5.2 Trial Enrollment Guidance

Always:

• Visit ClinicalTrials.gov for current trial status
• Contact trial sites directly to verify recruitment status
• Confirm eligibility criteria with principal investigators
• Verify IRB approval and informed consent requirements
• Consult with healthcare providers before enrolling

5.3 Trial Quality Disclaimer

Zeta does NOT evaluate or endorse clinical trials. We provide trial listings for research purposes only. Trial inclusion in our database does not imply:

• Safety or efficacy of investigational treatments
• Quality of trial design or execution
• Suitability for specific patients
• Lantern Pharma endorsement or affiliation

6. Knowledge Graphs

6.1 Relationship Visualization

Knowledge graphs are automatically generated from conversation content. They are:

• Simplified Representations: Complex relationships reduced to nodes and edges
• Not Comprehensive: May omit nuanced relationships or context
• Subject to Extraction Errors: Entity recognition may misidentify concepts
• Research Aids: Visual tools for exploration, not authoritative knowledge maps

Knowledge graphs are starting points for research, not definitive truth.

6.2 Relationship Strength Caveats

Edge weights (relationship strengths) are based on citation frequency, which may NOT reflect:

• Quality of evidence (high citation count ≠ strong evidence)
• Causation vs correlation
• Negative results (less frequently published)
• Recent breakthroughs (not yet highly cited)

7. Experimental Features (Beta)

7.1 Beta Software Status

• Features may change without notice
• Performance may vary significantly
• No uptime guarantee
• Data loss may occur (backup important work)

7.2 Ether0 Molecular Design (Experimental)

Ether0 is a 24-billion parameter chemistry AI model for molecular design:

• Status: Research-grade, not production-ready
• Output Quality: May generate non-synthesizable structures
• Validation Required: All designs require expert chemistry review
• No Guarantees: Generated molecules may not have desired properties

Ether0 outputs are starting points for medicinal chemistry, not synthesis-ready designs.

7.3 File Upload Feature (Future)

When file upload is enabled, uploaded documents will be processed by AI. Be aware:

• AI may misinterpret document content
• OCR errors may occur in scanned documents
• Complex tables or figures may not be accurately extracted
• Do NOT upload PHI or patient-identifiable information

8. No Liability for Decisions

• Medical Decisions: Patient diagnoses, treatment selections, or care decisions based on Zeta outputs
• Research Directions: Drug discovery strategies, target selection, or experimental designs informed by Zeta
• Financial Losses: Investment decisions, grant applications, or business strategies based on Zeta analysis
• Patient Outcomes: Harm resulting from medical decisions informed by Zeta outputs
• Errors & Omissions: Factual inaccuracies, missing information, or AI hallucinations
• Third-Party Data: Errors in PubMed, ORPHANET, or other external databases
• Opportunity Costs: Missed research opportunities or alternative approaches not suggested by Zeta

YOU ASSUME ALL RISK when using Zeta outputs for any purpose.

See our Terms of Service for complete liability limitations.

9. Reporting Errors

9.1 Safety Reporting

If you identify factual errors, dangerous outputs, or harmful recommendations:

Email: contact@withzeta.ai
Subject: "Zeta Safety Report"
Include:

• Your query (exact text)
• Zeta's response (copy/paste or screenshot)
• Explanation of the error or concern
• Correct information (if known) with citations

Response Time: We review all safety reports within 48 hours and will take corrective action if warranted.

9.2 Bug Reports

For technical bugs or system errors (not medical accuracy issues):

Email: contact@withzeta.ai
Subject: "Bug Report"

9.3 Community Responsibility

Beta users play a critical role in improving Zeta's safety and accuracy. We encourage you to report errors, especially those that could lead to patient harm or misguided research.

10. Regulatory Status

10.1 FDA Classification

10.2 HIPAA Compliance

withZeta is NOT HIPAA-compliant. Do NOT input:

• Patient names, dates of birth, or contact information
• Medical record numbers or health insurance information
• Social Security Numbers or government identifiers
• Diagnostic codes or treatment records linked to individuals
• Any data that could identify a specific patient

Violating HIPAA by inputting PHI into Zeta is YOUR responsibility. Lantern Pharma is not liable for HIPAA violations by users.

10.3 International Regulatory Status

Zeta is not approved or cleared by regulatory agencies outside the United States (e.g., EMA, PMDA, MHRA). Use in other jurisdictions may be subject to local laws and regulations.

10.4 No Medical Licensure

Lantern Pharma is not a licensed healthcare provider. Zeta does not practice medicine. All outputs are for research purposes only.

Last Updated: January 16, 2026

Version 1.0 (Beta)