Recently gave a talk about how it all works at HOPE 2026 and a bunch of folks talked to me afterwards inspiring me to share this technology more https://schedule.hope.net/hope26/talk/P9VFEH/
slides are up as well in case folks are interested, and the talk should eventually be placed online by the HOPE organizers
https://bit.ly/4hv295g
Completely understand, the legal landscape has really shifted around AI/LLM tools. I see tools drop everyday that spit in the face of DMCA/Copyright law but they skirt by mainly because they leverage AI
That has nothing to do with LLMs or not. If they were to "code" the same tool with an LLM, or build a LLM version of it, in Germany (where rthey are residing) they would face the same consequences.
Our legal system in that regard is total BS, when it comes to the so called "Hacker Paragraph". This comes with a sentence tht can range from a fine to up to one year prison time, just for producing such a tool. So if the police would actually - even if they had not published it - computer... well. They could be potentially "f**ed" (depending on how a judge would decide).
Agreed. Also less noticeable when "accidentally" dropped/left, especially in tight spaces out of view. And better odds of plausible deniability if caught or device is later found and somehow traced to you.
I built Nightcrawler, an open-source autonomous penetration-testing agent that runs entirely on an Android phone.
The project started with a question: how much of a real pentesting workflow could I run locally on relatively old mobile hardware, without relying on a cloud model or API?
Nightcrawler runs a 1.2B-parameter model locally on the Adreno GPU of a OnePlus 8. The model chooses targets and tools, while a separate scope-enforcement proxy validates every command before execution. The system maintains per-host memory in SQLite, rotates between targets, matches detected versions against a local CVE database, executes multi-step playbooks, and generates a structured report.
A few implementation details that may be interesting:
Local inference runs at roughly 115 prompt tokens/sec and 13 generated tokens/sec.
The small model only produces a usable command around 50% of the time, so much of the engineering is recovery logic, duplicate detection, persistent memory, and deterministic playbooks.
Every command passes through a separate scope and safety layer rather than trusting the model to remain in scope.
The project includes a dry-run mode, so the agent loop can be tested without executing real network commands or owning the phone hardware.
I've had it running on my home network for the past 3 months uninterrupted
So far only against four authorized networks. One of them was a corporate network. Left it overnight and it only found one minor week old CVE that I'm sure the IT team already had on their tracker. But showed that the proof of concept worked.
slides are up as well in case folks are interested, and the talk should eventually be placed online by the HOPE organizers https://bit.ly/4hv295g