Tool Receipts, Not Zero-Knowledge Proofs: Practical Hallucination Detection for AI Agents

Researchers propose NabaOS, a lightweight verification framework that detects AI agent hallucinations using HMAC-signed tool receipts instead of zero-knowledge proofs. The system achieves 94.2% detection accuracy with <15ms verification time, compared to cryptographic approaches that require 180+ seconds per query.

• →NabaOS uses HMAC-signed tool execution receipts to verify AI agent claims in real-time with minimal overhead.
• →The framework detects over 90% of fabricated tool references and false claims while requiring less than 15ms per verification.
• →Zero-knowledge proof approaches like zkLLM provide near-perfect accuracy but are impractical for interactive agents due to 3-minute processing times.
• →The system classifies claims by epistemic source using principles from Indian philosophy to give users actionable trust signals.
• →NabaOS offers the best cost-latency-coverage trade-off for practical AI agent verification in interactive applications.