Daniel Hammon·Sep 29Creating a Malicious ML Model with a Keras Lambda Layer — Then Reverse Engineering ItI created a malicious ML classifier by injecting a code-execution payload into a Keras Lambda layer. I then walk through how a security…A response icon1A response icon1
Daniel Hammon·Sep 26System 1 Models Are Vulnerable To Flip TokensIn this research, I find that the system 1 model Laya is vulnerable to a model-evasion technique HiddenLayer calls EchoGram. In my testing…
Daniel Hammon·Jul 14Protect AI’s LLM-Guard Prompt Injection Guardrail BypassI was curious if I could leverage the EchoGram technique to find flip tokens in open-source classification models and found that Protect…A response icon1A response icon1
Daniel Hammon·May 29Shift AgenticShift Agentic is the practice of embedding security controls, guidance, and feedback loops directly into AI coding agent workflows so that…A response icon1A response icon1
Daniel Hammon·Jan 24Blind Boolean-Based Prompt InjectionIn this post, I introduce and demonstrate the attack method Blind Boolean-Based Prompt Injection (BBPI) which is a prompt injection…
Daniel Hammon·Oct 12, 2025What You May Be Missing With Your Phishing Simulation ProgramPhishing simulation technology, where security teams send phishing emails to organization team members to assess how susceptible the…