Why Data Integrity Is Critical for AI and Web 3.0

By Bruce Schneier and Davi Ottenheimer
Originally published in IEEE Spectrum

The Shift to Integrity-Centric Systems

The internet’s evolution from Web 1.0 to Web 3.0 marks a pivotal shift—from availability and confidentiality to data integrity as the cornerstone of digital trust. Tim Berners-Lee’s 2014 call for a “Magna Carta for the Web” envisioned a user-owned internet. Today, decentralized protocols like ActivityPub (used by Mastodon) and Berners-Lee’s Solid project are making this a reality by embedding cryptographic verification and governance into data ownership.

Why Integrity Matters Now

AI’s autonomy amplifies risks:

• AI agents now handle financial transactions, healthcare decisions, and infrastructure control.
• Integrity failures—like corrupted inputs or biased outputs—can cascade into real-world harm (e.g., Boeing 737 MAX crashes, ChatGPT data leaks).
• Four critical domains:
  1. Input integrity: Authenticating data sources (e.g., thwarting voice-clone scams).
  2. Processing integrity: Ensuring algorithms function correctly (e.g., Ariane 5 rocket explosion due to a data overflow).
  3. Storage integrity: Preventing corruption (e.g., SolarWinds hack).
  4. Contextual integrity: Respecting data-use norms (e.g., Midjourney’s biased image generation).

Building Trust in AI Systems

Schneier outlines a blueprint for “integrous” systems (a revived term for systems with integrity):

• Cryptographic verification: Digital signatures and decentralized identifiers.
• Compartmentalization: Isolating failures like a fireproof kitchen.
• Transparency: Clear data lineage and governance (e.g., Solid’s personal data pods).
• Regulation: Aligning with global AI safety laws.

Challenges Ahead

• **Performance trade-offs: Integrity checks can slow real-time AI.
• Quantum computing threatens current cryptography.