AI Agents Need Data Integrity

This essay contends that data integrity—not just confidentiality or availability—must anchor AI and Web 3.0 security, detailing four integrity dimensions (input, processing, storage, contextual) and advocating controls like cryptographic verification, formal methods, compartmentalization, monitoring, and governance. It explains how decentralized architectures and standards (e.g., ActivityPub, Solid, DIDs) can strengthen provenance and accountability, and illustrates risks through historical failures (Ariane 5, Mars Climate Orbiter, SolarWinds, CrowdStrike, prompt injection, voice-clone scams). The authors call for a mindset shift, clearer language (reviving “integrous”), and aligned regulation to ensure trustworthy, resilient AI systems operating with reduced human oversight.