CrewCheck Journal
All Posts – Page 4
Accountability Metrics for DPDP AI Programs
DPDP operating note 30: a practical note for operators on activity metrics look healthy while compliance weakens.
3 Apr 2026 · 5 min read
Aadhaar Detection Needs More Than Regex
AI safety operating note 1: a practical note for AI platform engineers on plain regex catches invoice numbers and misses obfuscated Aadhaar text.
2 Apr 2026 · 5 min read
Output Scanning for PII Leakage
AI safety operating note 2: a practical note for LLM application teams on models can reintroduce personal data even after input redaction.
1 Apr 2026 · 5 min read
Prompt Injection Controls for Compliance Agents
AI safety operating note 3: a practical note for security teams on malicious pages can ask an agent to ignore privacy policy.
31 Mar 2026 · 5 min read
Trust Scores for AI Agents
AI safety operating note 4: a practical note for AI governance leads on binary pass/fail hides slow agent degradation.
30 Mar 2026 · 5 min read
Shadow AI Detection on Public Websites
AI safety operating note 5: a practical note for risk teams on marketing scripts quietly call model APIs outside governance.
29 Mar 2026 · 5 min read
Circuit Breakers for Unsafe AI Behavior
AI safety operating note 6: a practical note for SRE teams on unsafe agents continue serving traffic after repeated violations.
28 Mar 2026 · 5 min read
Safe Prompt Templates for Regulated Teams
AI safety operating note 7: a practical note for developer-experience teams on free-form prompts drift away from approved policy.
27 Mar 2026 · 5 min read
Human Review Queues for High-Risk AI Calls
AI safety operating note 8: a practical note for operations managers on some requests should pause instead of being auto-answered.
26 Mar 2026 · 5 min read
AI Safety Regression Tests for Indian PII
AI safety operating note 9: a practical note for QA leaders on a scanner update can break Aadhaar or PAN detection silently.
25 Mar 2026 · 5 min read
False Positives in PII Redaction
AI safety operating note 10: a practical note for product teams on over-redaction makes AI answers useless.
24 Mar 2026 · 5 min read
Multilayer PII Defense for LLM Gateways
AI safety operating note 11: a practical note for security architects on single-detector systems fail on formatting tricks.
23 Mar 2026 · 5 min read