Why autonomous reverse engineering needs a disprover
An agent that can only look for problems will always find them. The fix is not a better prompt: it is a second agent whose job is to fail the first.
Kleydin Research
Notes from the bench.
How the agents work, where they fail, and how we measure the difference. Design notes from the team building Kleydin, written before launch.
Rebuilding request signing as a Burp extension, automatically
Proxying a modern mobile app is easy. Editing a request and having the server still accept it is where a day disappears. That part is mechanical enough to automate.
Kleydin Research
How we benchmark detection and false alarms
A detection rate without a false-alarm rate is marketing. Here is how we measure both, what we count, and what we deliberately leave out.
Kleydin Research