Skip to content

The Red List: Data Security Protocol

NEVER SHARE THESE WITH PUBLIC AI MODELS

The Red List defines data types that must NEVER be uploaded to external AI services (ChatGPT, Claude, Gemini, etc.)

For sensitive data: Use local sandbox ONLY

Category 1: Unpublished Research Data

🚫 Novel Molecular Structures

Examples: - Chemical formulas for new compounds - Molecular structures before patent filing - Novel synthesis routes - Proprietary modifications to known structures

Why protected: Patent priority, competitive advantage

Solution: Use local sandbox with anonymised structures like [Compound A], [Novel polymer X]

🚫 Exact Synthesis Parameters

Examples: - Precise temperature/pressure conditions - Exact precursor ratios - Specific catalyst concentrations - Processing times and sequences - Equipment-specific settings

Why protected: Manufacturing know-how, competitive advantage

Solution: Use generic placeholders: [TEMP]°C, [X]% loading, [Y] hours

🚫 Experimental Results (Ongoing Research)

Examples: - Raw data from current experiments - Performance metrics before publication - Characterisation results - Yield data - Comparative test results

Why protected: Competitive advantage, patent disclosure risk

Solution: Wait until published OR use aggregate/anonymised data: "Yield improved ~20%" instead of "Yield: 87.3%"

🚫 Failed Experiments (Negative Data)

Examples: - Approaches that didn't work - Conditions that led to failure - Unexpected results

Why protected: Valuable for IP strategy (knows what doesn't work), can guide competitors

Solution: Use hypothetical framing: "What factors might cause [generic problem]?" instead of sharing specific failures

🚫 Grant Applications Under Review

Examples: - Unpublished research proposals - Preliminary data submitted with grants - Project timelines and milestones - Budget information

Why protected: Competitive funding advantage, IP disclosure

Solution: Wait until funded OR use public information only

Category 2: Commercially Sensitive Information

🚫 Exact Formulations

Examples: - Precise ingredient/component ratios - Additive identities and concentrations - Proprietary blends - Trade secret compositions

Why protected: Core competitive advantage

Solution: Use generic descriptions: "Polymer matrix with nano-reinforcement in optimal ratio"

🚫 Process Temperatures/Pressures/Conditions

Examples: - Exact processing windows - Optimised parameter sets - Equipment-specific conditions - Multi-step sequences with timing

Why protected: Manufacturing efficiency, quality control

Solution: Use ranges or categories: "Elevated temperature", "Moderate pressure"

🚫 Yield Data Revealing Efficiency

Examples: - Production yields - Waste percentages - Batch success rates - Process efficiency metrics

Why protected: Reveals manufacturing capability and costs

Solution: Use qualitative terms: "High yield", "Commercial viability demonstrated"

🚫 Cost Breakdowns

Examples: - Raw material costs - Processing costs per unit - Profit margins - Pricing strategies

Why protected: Competitive pricing advantage

Solution: Never share; if needed, use cost-effectiveness comparisons without absolute values

🚫 Customer/Partner Identities

Examples: - Customer names - Partner company identities - Collaboration details - Contract terms

Why protected: Confidentiality agreements, competitive intelligence

Solution: Use generic terms: "Aerospace partner", "European customer"

Category 3: Personal & Confidential Data

🚫 Employee Information

Examples: - Names with roles - Contact details - Salary information - Performance reviews - Personal data of any kind

Why protected: Privacy laws (GDPR, etc.), professional ethics

Solution: Never share; use roles only: "Senior Researcher", "Lab Technician"

🚫 Customer Data

Examples: - Contact information - Order details - Communication history - Technical requirements - Company-specific information

Why protected: Confidentiality agreements, data protection law

Solution: Never share; if needed for templates, use completely anonymised examples

🚫 Internal Communications with Strategy

Examples: - Strategy meeting notes - Email discussions about direction - Decision-making rationale - Internal debates

Why protected: Competitive strategy

Solution: Never share; for formatting help, use sanitised examples

🚫 Financial Data

Examples: - Revenue figures - Budget allocations - Forecasts - Bank details - Investment information

Why protected: Competitive intelligence, security

Solution: Never share

Examples: - Contracts - NDAs - IP correspondence - Litigation materials - Legal advice

Why protected: Attorney-client privilege, confidentiality

Solution: Never share; seek legal counsel for advice

Category 4: Security-Sensitive Information

🚫 Access Credentials

Examples: - Passwords - API keys - Authentication tokens - System access details

Why protected: Security breach risk

Solution: NEVER share; rotate immediately if accidentally exposed

🚫 System Configurations

Examples: - Server setups - Network architecture - Software configurations - Database schemas

Why protected: Security vulnerability

Solution: Never share

🚫 Security Protocols

Examples: - Physical access procedures - Data backup systems - Incident response plans - Vulnerability assessments

Why protected: Security effectiveness

Solution: Never share

Decision Tree: Can I Share This?

START: I want to use AI for [task]

↓

Does the task involve Red List data?
├─ NO → Proceed, but verify output
└─ YES → Can I sanitise the data?
    ├─ YES → Sanitise, then use external AI
    │   └─ Verify sanitisation is complete
    │       - All specific values removed?
    │       - All identities anonymised?
    │       - No reverse-engineering possible?
    │
    └─ NO → Use local sandbox ONLY
        └─ Benefits:
            - Complete privacy
            - No external logging
            - Full control

Sanitisation Techniques

Technique 1: Placeholder Replacement

Before (Red List violation): 

"Synthesised at 750°C for 4 hours using 2.5% graphene oxide"

After (Safe): 

"Synthesised at [TEMP]°C for [TIME] hours using [X]% nanofiller"

Technique 2: Aggregation

Before (Red List violation): 

"Sample A: 45.2 MPa, Sample B: 47.1 MPa, Sample C: 44.8 MPa"

After (Safe): 

"Tensile strength averaged 45-47 MPa across samples"

Technique 3: Hypothetical Framing

Before (Red List violation): 

"Our novel PLA/graphene composite with in-situ reduction shows 40% improvement"

After (Safe): 

"For polymer/nanofiller composites generally, what factors typically influence mechanical property improvements of 30-50%?"

Technique 4: Generic Domain Transfer

Before (Red List violation): 

"Need to optimise electrospinning of PLA at 22kV for customer X"

After (Safe): 

"General principles for optimising electrospinning parameters for thermoplastic polymers?"

Red List Violation Examples

❌ Example 1: Unpublished Data

Prompt: "Analyse this data from Experiment #343. Samples showed 
tensile strength of 42.3, 45.1, 44.8 MPa at 5% graphene loading 
using our proprietary in-situ reduction method at 95°C for 3 hours..."

Violations: - Exact experimental data (unpublished) - Specific synthesis conditions - Proprietary method details

Fix: Use local sandbox OR wait until published

❌ Example 2: Customer Information

Prompt: "Draft email to Aerospace Innovations Ltd about 3-week delay 
in their 50kg order of ceramic nanofibers. They're our largest 
customer (40% revenue). Need to maintain relationship..."

Violations: - Customer name - Order details - Revenue information

Fix: 

"Draft email template for customer communication regarding production 
delay. Context: Important client, technical issue, maintaining 
relationship. Use generic placeholders."

❌ Example 3: IP Disclosure

Prompt: "Review this patent draft for our novel [detailed invention 
description before filing]..."

Violations: - Pre-filing disclosure (affects patent priority) - Complete invention details

Fix: Use local sandbox OR wait until filed

Incident Response

If you accidentally share Red List data:

  1. STOP: Do not continue the conversation
  2. DOCUMENT: Screenshot what was shared, timestamp
  3. REPORT: Immediately notify:
  4. Your supervisor
  5. IT/Security team
  6. Legal (if IP-related)
  7. ASSESS: Determine sensitivity and potential impact
  8. MITIGATE: Follow company incident response protocol
  9. LEARN: Update processes to prevent recurrence

Do NOT: - Try to "delete" the conversation (data may be logged) - Continue using that session - Assume "no harm done"

Training Verification

Test your understanding:

Scenario 1

"I want to format 20 synthesis protocols into standard templates. The protocols contain exact temperatures, pressures, and precursor ratios for our proprietary process."

Assessment: ☐ Green ☐ Yellow ☐ Red
Action: _____

Scenario 2

"I want to summarise 5 published papers on PLA degradation (all with DOIs, published 2020-2023)."

Assessment: ☐ Green ☐ Yellow ☐ Red
Action: _____

Scenario 3

"I want to analyse SEM images from our patent-pending process, filename contains 'AmaDema_Proprietary_Method_v7'."

Assessment: ☐ Green ☐ Yellow ☐ Red
Action: _____

Answers: 1. Red - Use local sandbox (proprietary process parameters) 2. Green - Proceed (published data) 3. Red - Use local sandbox (patent-pending process)

Remember: When in doubt, use local sandbox. Privacy is permanent; exposure is irreversible.

Next: External Resources → Summary