LLM Security
-
Prompt Injection Mitigation: What Holds and What Breaks
Prompt injection mitigation ranked by evidence: which architectural controls cap blast radius and which filters adaptive attacks already broke.
-
Prompt Injection in AI Agents Explained
Agents turn prompt injection into tool calls, code execution and data theft. How the attack works when a model can act, and what contains it.
-
Indirect Prompt Injection: How These Attacks Work
Indirect prompt injection hides instructions in content the model retrieves. The delivery channels, the attacker objectives, and what benchmarks measure.
-
Prompt Injection Testing: A Repeatable Method
Prompt injection testing needs a method, not just a scanner. How to scope surfaces, pick a corpus and harness, define detectors, and score attack success.
-
Prompt Injection Detection Techniques That Actually Work
Compare prompt injection detection techniques: pattern filters, fine-tuned classifiers, and known-answer probes, plus why no single layer closes the hole.
-
How to Mitigate Prompt Injection: A Layered Defense Guide
No single control stops prompt injection. This guide covers the layered stack that works: input and output filtering, privilege limits, and isolation.
-
OWASP LLM Top 10 Prompt Injection (LLM01:2025) Explained
LLM01 in the OWASP LLM Top 10 is prompt injection — and it held the top spot in both the 2023 and 2025 editions. What the 2025 entry actually says.
-
Invisible Prompt Injection via Unicode Tag Smuggling
Unicode Tag characters let attackers embed invisible prompt injection payloads that still tokenize as instructions. How it works and what stops it.
-
Garak vs PyRIT vs promptmap: Prompt Injection Testing
Garak, PyRIT, and promptmap are not interchangeable. What each prompt injection testing framework is built for, what it misses, and how to choose one.
-
Rebuff Prompt Injection Detector: How It Works and Fails
Rebuff, Protect AI's prompt injection defense library, was archived on GitHub in May 2025. How its four detection layers work and what each one misses.