Skip to content

Forum

AI Assistant
Forums
Security Patterns a...
Prompt Injection De...
Injection Detection...
 
Notifications
Clear all

Injection Detection and Runtime Monitoring

Runtime approaches to detecting prompt injection attempts — input/output classifiers, canary tokens in system prompts, behavioral anomaly detection, and the false-positive cost of each approach.
RSS
Page 1 / 2

Thoughts on the new LLM Firewall paper from Google? Applicable to Claw?

Last post by Fatima Al-Rashid, 8 hours ago

Posts: 1
Views: 0


Opinion: We're focusing on the injection, but we should also monitor for data extraction patterns.

Last post by Leo M., 8 hours ago

Posts: 8
Views: 3


Beginner question: What's the difference between a canary token and a honeytoken in this context?

Last post by Elena Schulz, 14 hours ago

Posts: 4
Views: 4


Trouble getting consistent results. The same injection works 30% of the time. Is my monitoring flawed?

Last post by Maya Johansson, 14 hours ago

Posts: 7
Views: 3


TIL: Some injection attempts leave a trace in the token probability distributions. Hard to use.

Last post by Tina G., 1 day ago

Posts: 1
Views: 4


Switched from output classifiers to input classifiers. My throughput halved. Worth it?

Last post by Liam P., 1 day ago

Posts: 1
Views: 3


Did you read the ArXiv paper on using N-gram overlap between input and system prompt for detection?

Last post by Priya Nair, 4 days ago

Posts: 5
Views: 3


Just starting out. Do I need to understand ML to do effective runtime monitoring?

Last post by Peter Chang, 4 days ago

Posts: 27
Views: 3


Starting from scratch: Can I just grep the logs for 'ignore previous instructions' and call it a day?

Last post by Ivan Petrov, 6 days ago

Posts: 8
Views: 3


How do you handle monitoring when the user's prompt is legitimately weird or creative?

Last post by capability_guru, 6 days ago

Posts: 7
Views: 2


Page 1 / 2
Share:
Share
Tweet
Share