14 defence instruction defence

ScottLogic / prompt-injection

Application which investigates defensive measures against prompt injection attacks on an LLM, with a focus on the exposure of external tools.

MIT License

15 stars 10 forks source link

14 defence instruction defence #760

Closed heatherlogan-scottlogic closed 8 months ago

heatherlogan-scottlogic commented 8 months ago

Description

Instruction defence from https://learnprompting.org/docs/prompt_hacking/defensive_measures/instruction

Screenshots

Notes

Bullet point list with
any notable code changes
or explanations regarding this PR

Concerns

Bullet point list
with any questions
or concerns about this PR

Checklist

Have you done the following?

[x] Linked the relevant Issue
[x] Added tests
[x] Ensured the workflow steps are passing

heatherlogan-scottlogic commented 8 months ago

Code all looks good, and the defence seems to work correctly on testing, apart from a seperate bug (which is in dev as well) which I've made a ticket for: #762

that bug will be fixed with https://github.com/ScottLogic/prompt-injection/pull/753