promptexploit
i'm feeling
★ adversarial ★
SYS.NAME · promptexploit
SYS.AUTH · GUEST_OK
SYS.NODE · promptexploit.com
UPTIME · 142 days
TERMINAL · TTY0
STATUS · 200 OK
$ whoami
Breaking and defending LLMs. Notes on prompt injection, jailbreak research, and agent security — mostly so the defenses get better.
$ ls -la /posts/
-
LLM04
llm04-2025-data-and-model-poisoning-simple-explanation.mda simple explanation of data and model poisoning in LLM apps
-
LLM03
llm03-2025-supply-chain-simple-explanation.mda simple explanation of supply chain risk in LLM apps
-
LLM02
llm02-2025-sensitive-information-disclosure-simple-explanation.mda simple explanation of sensitive information disclosure in LLM apps
-
PI
llm01-2025-prompt-injection-simple-explanation.mda simple explanation of OWASP LLM01 prompt injection
-
OWASP
owasp-genai-llm-top-10-simple-explanation.mda plain-language map of the GenAI Security Project and LLM Top 10
-
PI
indirect-prompt-injection-101.mdhow untrusted content smuggles instructions into a model
-
DEF
isolating-tool-output-in-agents.mdpatterns that stop agents from trusting their inputs blindly
-
RT
building-a-jailbreak-eval-harness.mdmeasuring model robustness with a repeatable test suite