Uncensored Ai for developing hacking tools

Discussion & Solutions
1

I am looking for AI who help me to develop Hacking tool like Rats, Reverse Engineering, Bug finding ,etc. Anyone know?

2

A RAT is malware, so no chat model will write one and no amount of prompting changes that. What is open, enormous and free is the literature those tools are assembled from — and in 2026 there are local models and agents built specifically to read it with you. That is the part nobody in this space talks about, and it is the only part that actually makes you better. :wrench:

digraph G { rankdir=TB; bgcolor="transparent"; nodesep=0.32; ranksep=0.40; node [shape=box style="rounded,filled" fontname="Helvetica" fontsize=11 color="#2E4A6B" fillcolor="#EDF3F8" penwidth=1.3]; edge [color="#8A9A8F" penwidth=1.3 arrowsize=0.8]; ASK [label="your ask: an AI that helps me build\nand break things" fillcolor="#FDF3E7" color="#8C5B3B"]; LIB [label="📚 the literature\nPayloadsAllTheThings · GTFOBins\nLOLBAS · HackTricks · Vulnhuntr" fillcolor="#FBF7E8" color="#A08A3C"]; MOD [label="🧠 the model\nWhiteRabbitNeo · Dolphin · Hermes\nabliterated coder, running locally" fillcolor="#F3EAF7" color="#6B4E8C"]; AGT [label="🤖 the agent\nStrix · PentestGPT · CAI\nButtercup · hackingBuddyGPT"]; MCP [label="🔬 the eyes\nGhidraMCP reads your Ghidra project"]; OUT [label="the loop you actually want\nask a technique -> it greps the literature\n-> shows the line -> explains why" fillcolor="#EAF5EC" color="#2F6B4A"]; LIB -> OUT; MOD -> OUT; AGT -> OUT; MCP -> AGT; ASK -> LIB; ASK -> MOD; ASK -> AGT; ASK -> MCP; }

:world_map: Your three asks, one lane each

reverse engineering   ->  GhidraMCP          (an AI reading your Ghidra project)
bug finding           ->  Strix + Vulnhuntr  (an agent that breaks apps and explains where)
tooling and payloads  ->  the four repos     (every technique, in plain text, greppable)

:books: The four bibles, and what each one gives you

  • PayloadsAllTheThings ★81,985 — every payload, bypass and trick for web security, by category, with the exact string to paste
  • GTFOBins ★13,780 — the Unix privilege escalation one-liners, ready to copy
  • LOLBAS ★8,899 — the same idea for Windows: living-off-the-land binaries and scripts
  • HackTricks — the field manual that links all of the above into an order of operations
  • Vulnhuntr ★2,806 · AGPL — zero-shot bug hunting: point it at a repo, get the call path and the reason

Every one is a git repo, which means an agent can search them better than any search engine: grep the technique, not the blog.

:brain: The models that will actually discuss the technique

  • WhiteRabbitNeo — a model family trained for offensive and defensive security, tuned to answer technique questions without the lecture
  • Dolphin and Hermes — general uncensored weights that hold up on exploit theory
  • An abliterated coder — abliterated means the refusals were cut out of the model file, so it explains the assembly it just wrote
  • Run them locally with ollama: a 30B mixture-of-experts fits in 32 GB of RAM, answers fast on a CPU, and logs nothing anywhere

:robot: The agents that do the work

  • Strix ★71,232 · Apache-2.0 — finds and fixes vulnerabilities in a running app, open source, pushed today
  • PentestGPT ★16,081 · MIT — plans a pentest and keeps an auditable reasoning trail
  • CAI ★9,840 — humans and agents in one security team, recon through reporting
  • Buttercup ★1,727 · AGPL — from Trail of Bits, and it patches what it finds
  • hackingBuddyGPT ★1,282 · MIT — the entire concept in fifty readable lines, which is the fastest way to see how an agentic loop actually works

:microscope: Reverse engineering, the skill that compounds

  • GhidraMCP ★11,489 — an MCP server that lets an AI read your Ghidra reverse-engineering project — rename functions, follow cross-references, explain a decompiled block in place. The single biggest upgrade to RE since Ghidra itself
  • ghidra-mcp ★5,855 — 200+ tools exposed
The combination that beats any single tool

Clone the four repos, run one uncensored model locally, and point an agent at both. Ask it for a technique, not a payload: “privilege escalation on a Linux box with rsync installed and no sudo”. It answers from GTFOBins, shows the line, and explains why it works — with no server watching, because none of it left your machine.

That is the real capability jump: agent + scripture + a local model, on your own hardware, for zero currency. The technique is the asset — persistence, tunneling, payload delivery, evasion — and all of it lives in those four repos in plain text.

Build the loop before the tool: literature in, model local, agent reading both. The tool is the last thing you write, and by then you will know exactly why it works. :test_tube:

1 Like