AI Summaries Can Rewrite the Record

A summary selects, compresses and interprets. When it replaces the source, the institution may inherit the model’s judgment as fact.

The answer

The AI summary is easier to read than the record. It is therefore more likely to be remembered, forwarded and cited. Compression removes qualification, dissent and sequence. A generated sentence can become the institution’s history.

The AI summary is easier to read than the record. It is therefore more likely to be remembered, forwarded and cited.

Compression removes qualification, dissent and sequence. A generated sentence can become the institution’s history.

Separate record, interpretation and decision

Label summaries as derived. Link exact sources and effective dates. Preserve uncertainty and material disagreement.

Value object — The Summary Provenance Block

- Source records and versions.

- Purpose and audience.

- Material omissions.

- Inferences added.

- Reviewer corrections.

- Expiry when sources change.

Never make the summary the only copy

Use it as a view into the record. For legal authority, ownership and consequential decisions, require direct source inspection.

A summary saves attention. It should not acquire the power to rewrite what happened.

Where this breaks

Summaries become dangerous when copied into board packs, CRM records or memory without a link to the source. Later users cannot distinguish the model’s compression from established fact.

The operating move

Treat consequential summaries as governed derived records. Include provenance, omissions, uncertainty and a refresh trigger tied to source change.

Cite exact records and versions.

Separate fact from inference.

Preserve material dissent.

Expire summaries after change.

The test

Correct a source fact and trace whether every summary, embedding and downstream record updates. Any surviving old conclusion is active misinformation.

Sources

  1. NIST: AI Risk Management Framework — Generative AI ProfileNIST: AI Risk Management Framework

    Primary authority

  2. Swiss FDPIC: AI and data protectionSwiss FDPIC: AI and data protection

    Primary authority

  3. NIST SP 800-207: Zero Trust ArchitectureNIST SP 800-207: Zero Trust Architecture

    Primary authority

Adam J. De CollibusFounding Partner, Svperior / Systems Engineering

Ross Belhomme

Ross leads Legal within Svperior GmbH. His work draws on more than two decades across international fiduciary, wealth-structuring, and private-client environments, combining legal, financial, and technical judgment around governance, privacy, assets, authority, and cross-border operating conditions.

Legal strategy / Governance / Private-client structuring / Digital assets

Need to apply this to a specific situation?

Send us the initial context. If the matter fits, we will respond directly.

Send private inquiry
AI Summaries Can Rewrite the Record \