Threat models for what runs on its own
Every collector and automation has a published threat model.
Every collector and automation has a published threat model.
What we do
- Write a threat model per collector, scheduled workflow and drop: what it reads, writes and trusts.
- List what could go wrong and the mitigations.
- Have the security agent review each one.
- Run a prompt-injection pass at least daily.
What we never do
- Never let an automation run without a threat model.
- Never act on instructions found in fetched text.
How to check us
In Dinis's words: 2026-11-18-inner-passes-and-security.md. Drafted from his guidance; journalist@ revises the text.
Licences on this page
What PrompTugal.news wrote and made here: CC BY 4.0.
Everything else linked here (1 other sources) stays under each source's own terms: follow the item's link to its source. We never assume a source is open.
How we treat licences: Licences.