Leak Cost Estimator
What Leaks in a Single Prompt
A typical developer prompt to an LLM contains more sensitive data than most teams realize. Here's what commonly appears — without anyone noticing.
API Keys & Tokens
AWS access keys, GitHub PATs, Stripe secrets — copied from .env files into prompts during debugging. A single leaked AWS key can pivot into full infrastructure access.
Source Code
Entire functions, proprietary algorithms, and architecture decisions pasted into prompts for review. Samsung banned ChatGPT after employees leaked source code three separate times in 2023.
Customer PII
Names, emails, phone numbers, and addresses from support tickets copied into AI tools. GDPR fines can reach 4% of global annual revenue per incident.
Health Records (PHI)
Patient data, diagnoses, and treatment plans. HIPAA penalties start at $50K per violation category and scale to $1.5M/year — even for accidental exposure through an LLM.
Financial Data
Account numbers, transaction records, investment strategies. PCI DSS non-compliance fines start at $5K/month, and financial institutions face additional SOX exposure.
System Prompts & Architecture
Internal system prompts that reveal product design, business logic, and competitive strategy. Once logged by a provider, these become discoverable by anyone with admin access.
The Cost Multiplier Effect
An AI data leak doesn't cost you once — it compounds across three stages as detection, legal, and regulatory machinery kicks in. The average breach takes 204 days to identify and 73 days to contain (IBM 2024).
Immediate Response
- Incident response team activation
- Forensic investigation begins
- Internal communication crisis
- Initial containment measures
Legal & Notification
- Outside counsel retained
- Regulatory notification (GDPR: 72hr deadline)
- Customer notification campaign
- Credit monitoring for affected users
Regulatory & Remediation
- Regulatory investigation & fines
- Customer churn & lost business
- System architecture remediation
- PR / reputation management
Real-World Scenarios
Three composite scenarios based on actual AI leak incident patterns. Names and details are illustrative but the cost structures are drawn from IBM Ponemon data, GDPR/SOC 2 enforcement actions, and public AI incident disclosures.
The API Key Cascade
What happened: A 30-person Series A startup. An engineer debugging an integration copies their AWS_ACCESS_KEY_ID into a prompt. The provider logs the request. Two weeks later, the provider's log storage bucket is misconfigured — the key is exposed in a public S3 bucket discovered by a security researcher.
The damage: The key granted IAM admin access. The attacker spun up $45K in crypto mining instances before detection. The startup spent $78K on forensic investigation, $12K on legal counsel, and lost a $200K enterprise deal when the prospect's security team flagged the incident. Total: ~$335K — roughly 10x the cost of a Foundation tier Shield license.
The Customer Data Exposure
What happened: A 200-person fintech company processing $50M/month in transactions. A customer support agent pastes 10,000 customer support tickets into an LLM for sentiment analysis. Tickets contain full names, email addresses, partial account numbers, and transaction amounts. The LLM provider stores prompts for 90 days under their data retention policy.
The damage: Under GDPR, 10K records of EU customer PII triggers mandatory notification within 72 hours. The company faced €800K in GDPR fines (reduced from potential €2M due to cooperation), $250K in notification costs, $150K in legal fees, and 4% customer churn ($80K MRR loss). The PCI DSS assessment following the incident flagged additional gaps costing $90K to remediate. Total: ~$1.7M.
The PHI Breach
What happened: A 1,500-employee healthcare analytics company. A data scientist uploads 50,000 de-identified patient records to an LLM for research pattern extraction. The records were 'de-identified' using simple field removal — but the combination of zip code, age, procedure date, and diagnosis code allowed re-identification of 18,000 patients (a well-known vulnerability documented in HIPAA guidance).
The damage: OCR opened an investigation. 18,000 patients qualified as a HIPAA breach requiring individual notification. The company faced $1.2M in HIPAA civil monetary penalties (Tier 3: willful neglect, corrected within 30 days), $350K in OCR-mandated corrective action plan implementation, $200K in patient notification and credit monitoring, $180K in legal defense, and lost a $4M hospital system contract during the investigation. Total: ~$5.9M.
Stop the leak before it starts
PurfectShield runs on your machine — redacting PII, secrets, and proprietary data before it ever leaves your device. No cloud dependency, no provider trust required. One environment variable to configure, zero changes to your code.