When the AI red team brings its own Kali sandbox
NeuroSploit exists to automate offensive security assessments inside disposable Kali sandboxes while forcing LLMs to prove their own findings before reporting them.

What it does
NeuroSploit is an autonomous penetration-testing platform that spins up isolated Kali Linux Docker containers per scan and orchestrates LLM-driven reconnaissance, payload testing, and tool execution in parallel. A FastAPI backend and React frontend provide real-time monitoring, while the system catalogs 100 vulnerability types across 10 categories and attempts to chain findings into broader exploit paths. It routes requests through multiple LLM providers and generates per-finding proof-of-concept code with confidence scores.
The interesting bit
The project treats LLM unreliability as a first-class threat: a dedicated validation layer uses negative controls, proof-of-execution checks across 25-plus methods, and a sole validation_judge to reject hallucinated findings before they reach the report. It also adapts strategy mid-scan, detecting dead endpoints and recomputing priorities rather than burning tokens on barren ground.
Key highlights
- 3-stream parallel architecture splits work into reconnaissance, junior payload testing, and tool-running streams that feed into deep analysis
- Per-scan Kali sandboxing with 56 tool recipes and a container pool, isolating each engagement from the host and from other scans
- Anti-hallucination pipeline with benign-request negative controls, numeric 0–100 confidence scoring, and a hard validation gate
- Exploit chaining engine with 10 rules to connect findings like SSRF to internal services or SQLi to database-specific attacks
- Multi-provider LLM routing for Claude, GPT, Gemini, Ollama, LMStudio, and OpenRouter
Caveats
- The README’s quick-start clone URL points to
your-org/NeuroSploitv2.git, a mismatch with the v3 branding and repository location that suggests the docs may not be fully synchronized with the current release.
Verdict
Security teams drowning in manual reconnaissance and false-positive triage will find the architecture thought-provoking; shops requiring enterprise support or turnkey compliance should probably steer clear until the ecosystem matures.
Frequently asked
- What is JoasASantos/NeuroSploit?
- NeuroSploit exists to automate offensive security assessments inside disposable Kali sandboxes while forcing LLMs to prove their own findings before reporting them.
- Is NeuroSploit open source?
- Yes — JoasASantos/NeuroSploit is open source, released under the MIT license.
- What language is NeuroSploit written in?
- JoasASantos/NeuroSploit is primarily written in Rust.
- How popular is NeuroSploit?
- JoasASantos/NeuroSploit has 1.2k stars on GitHub.
- Where can I find NeuroSploit?
- JoasASantos/NeuroSploit is on GitHub at https://github.com/JoasASantos/NeuroSploit.