# The Lethal Trifecta

DevFeed: [The Lethal Trifecta](<https://devfeed.tech/articles/the-lethal-trifecta-100416.md>)

Original publisher: [Read original article](<https://blog.alexewerlof.com/p/lethal-trifecta>)

Author: Alex Ewerlöf

Published: 2026-10-26T05:30:43Z

Content type: article

Language: en

Sources: [Alex Ewerlof Notes](<https://devfeed.tech/sources/alex-ewerlof-notes.md>)

Topics: [prompt injection](<https://devfeed.tech/topics/prompt-injection.md>), [Security](<https://devfeed.tech/topics/security.md>), [least privilege](<https://devfeed.tech/topics/least-privilege.md>), [software-architecture](<https://devfeed.tech/topics/software-architecture.md>)

Tags: [attack](<https://devfeed.tech/tags/attack.md>), [detection](<https://devfeed.tech/tags/detection.md>), [lethal-trifecta](<https://devfeed.tech/tags/lethal-trifecta.md>), [mitigation](<https://devfeed.tech/tags/mitigation.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>)

## AI overview

Prompt injection becomes especially risky when an agent combines three capabilities: reading untrusted input, accessing private data, and sending or changing information through an egress path. The article recommends mapping those capabilities and their data flows, then reducing blast radius through least privilege and short-lived permissions. Its dual-agent pattern has one agent read untrusted content and return a validated, strict JSON request; a separate executor uses sensitive tools without seeing the raw input. This adds complexity and still depends on the reader not passing along an attack, and the article cautions that agents may find indirect paths around controls. Blocking egress can limit data leaks but does not prevent harmful changes to internal state.

## Source excerpt

Detection and mitigation of a common prompt injection attack