Paste any suspicious message to check if it matches known manipulation templates being circulated online to trick AI systems.
β οΈ WHAT IS PROMPT INJECTION?
Malicious actors share copy-paste prompts designed to trick AI systems like Claude and ChatGPT into producing harmful content β scam frameworks, get-rich-quick schemes, or bypassing safety systems. TrustShield scans text before it reaches an AI, flagging known attack patterns.
Verified NZ Business Directory
Every reviewer is identity-verified by TrustShield. Scores are size-adjusted β a small business competes fairly with a large one.
π HOW THE FAIR SCORE WORKS
Verified Review %
Only authenticated human reviewers count. 100% of Gold reviewers are ID-verified.
Size-Adjusted Baseline
A micro business with 8 perfect reviews scores comparably to a large chain with 200.
Consistency Bonus
Sustained quality over time. A bad month doesn't destroy years of good service.
Resolution Bonus
Businesses that resolve disputes fairly score higher. Transparency is rewarded.
β
Gold
100% verified
β
Silver
Mostly verified
β
Bronze
Entry level
How TrustShield Works
Two systems working together: an AI abuse prevention layer, and a business trust verification framework built on dialogue β not one-way broadcasts.
π
Prompt Injection Detection
Community-curated database of known manipulation prompt templates
Pattern matching on structure, not just keywords β evolves as new patterns emerge
Open source signature library so researchers can contribute new patterns
API available for AI platforms, schools, and businesses to integrate protection
NZ CERT partnership to escalate novel attack patterns to national cybersecurity
πͺ
Gold Star Verification Process
Applicants submit NZ Business Number and proof of operation
TrustShield recruits reviewers independently β the business never selects them
Each reviewer completes identity verification (NZ driver licence or passport)
Reviews are structured interviews, not free text β harder to fake or game
Minimum 20 verified reviewers for Gold Β· 10 for Silver Β· 5 for Bronze
π¬
Dialogue-First Resolution
Unlike Trustpilot, reviews are not one-way broadcasts β both parties must engage
Business has 7 days to formally respond to any review within the platform
Disputed reviews go to a TrustShield mediator β both sides provide evidence
Reviewers who cannot substantiate claims have their review removed entirely
Resolved disputes are marked RESOLVED visibly β rewarding businesses who fix problems
Repeat bad-faith reviewers are permanently banned from the platform
π
Size-Fair Scoring Formula
Base Score = (Verified positive reviews Γ· Total verified reviews) Γ 100
Size bonus: Micro +8pts Β· Small +5pts Β· Medium +2pts Β· Large +0pts
Consistency bonus: 3+ years stable = +3pts Β· 1β3 years = +1pt
No business can buy a rating β all reviewer selection managed by TrustShield
π‘ CONCEPT NOTE
This is a product concept prototype built for New Zealand. The prompt injection scanner can be open-sourced immediately. The rating system would benefit from partnership with MBIE, Consumer NZ, and NZ CERT. Both systems address real, documented harms happening right now in New Zealand.