Research

Sixty-eight models agree, and none of it is the bug

sixty-eight open code models converge on a representation of vulnerable functions that does not contain the vulnerability

7 Sept 2026Maciej Cichoń

Reads the code but not the vulnerability

an unfitted reader of model activations cannot tell a vulnerable function from its own fix

15 Aug 2026Maciej Cichoń

Code written to be misread by an LLM

prompt injection, invisible unicode and adversarial rewrites aimed at the model reading your code

3 Aug 2026Maciej Cichoń

What Vulnerability Detectors Actually Learn

Testing whether code models encode vulnerabilities or just learn labels

20 Jul 2026Maciej Cichoń