Article / Private Capability

Prompt Injection Is an Authority Attack

Injected instructions matter because they attempt to redirect the authority given to a model or agent.

Written byJonathan P. De Collibus

Founding Partner, Svperior / Cyber

Published16 July 2026

DisciplinesCyber / Legal / Systems Engineering

The answer

A hostile document is not merely bad content. It tells the system to disclose, call a tool or ignore the user’s mandate. Treat retrieved instructions as untrusted data and enforce authority outside the model.

A hostile document is not merely bad content.

Where this breaks

It tells the system to disclose, call a tool or ignore the user’s mandate.

The operating move

Treat retrieved instructions as untrusted data and enforce authority outside the model.

Value object — The operational card

- Instruction source

- Permitted authority

- Tool boundary

- Detection evidence

The test

Place a hostile instruction in a test document. The agent should cite content without obeying it.

Evidence

Sources

NIST: AI Risk Management Framework — Generative AI ProfileNIST: AI Risk Management Framework
Primary authority
Swiss FDPIC: AI and data protectionSwiss FDPIC: AI and data protection
Primary authority
NIST SP 800-207: Zero Trust ArchitectureNIST SP 800-207: Zero Trust Architecture
Primary authority

Expert review

Adam J. De CollibusFounding Partner, Svperior / Systems Engineering

Author

Jonathan P. De Collibus

Jonathan co-founded Svperior in 2014 and leads its cyber practice. His work sits where adversarial pressure, technical architecture, and consequential decisions meet, with experience across clinical, financial, public-sector, and private-client systems where confidentiality, continuity, and technical correctness carry material consequences.

Cyber strategy / Adversarial assessment / Security architecture / Private systems

Private inquiry

Need to apply this to a specific situation?

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

Send private inquiry