Source: raw/These_10_Open_Source_Tools_Should_Be_Illegal.md — Sharbel (@sharbelxyz), “These 10 Open Source Tools Should Be Illegal” (youtube.com/watch?v=l_TDLJ97Dm0), YouTube auto-captions fetched 2026-10-30.
Creator: Sharbel · URL: https://www.youtube.com/watch?v=l_TDLJ97Dm0 · Duration: not recorded · Platform: YouTube
Scope of this article: awareness and defense only. It records what each tool is publicly known to do and the protective step Sharbel pairs with it. It gives no usage instructions, and the source gives none either. Sharbel’s own disclaimer: everything shown is legal “on your own devices and your own network”, and pointing any of it at someone else “you could be committing a crime.”
Sharbel’s video is a tour of ten free, popular GitHub projects. Six are classic security tools that show how visible a network, a person or a website already is. Four are AI tools that “manufacture” reality: a one-topic-to-finished-video pipeline, a 10-second voice cloner, an AI-label stripper, and an agent that reverse-engineers software. His closing point: none of these is hidden, so “the barrier was never access. The barrier was only ever knowing they exist.”
Key Takeaways
- Voice is no longer proof of identity. An open-source voice cloner (Chatterbox, from Resemble AI) copies a voice from about 10 seconds of audio in 23 languages. Sharbel: if your bank still uses a voice password, turn it off.
- The AI clone can be detected; the label-stripper is weaker than its reputation. Chatterbox stamps an inaudible watermark that survives compression, and a detector can prove a clip is synthetic. The stripper removes metadata and provenance tags, but not statistical watermarks, which its author acknowledges.
- AI content pipelines are now free to download, not free to run. A 110k-star project turns one topic into a finished short (script, stock footage, voiceover, subtitles), but each video still costs paid API calls and the output is “generic”. “It just prints filler” and shows “where half the videos in your feed already come from.”
- Your username is a map. A username search across 400+ sites finds every account that reuses the same handle in about 30 seconds. It “punishes consistency”, so run it on yourself once.
- Most of the classic attacks fail against basic hygiene: long Wi-Fi passwords on modern WPA3, HTTPS everywhere, patched websites, and refusing unexpected certificate or profile installs.
- The #1 tool is an AI agent doing the expert’s job. A reverse-engineering plugin for Claude Code or Cursor (most likely REA, see Reverse-Engineering Software With Coding Agents) ^[inferred] hit #1 on GitHub trending. It’s in beta, built by one developer, and inherits the model’s mistakes.
The Ten, as Exposures and Defenses
| # | Tool (what it is) | What it exposes, per Sharbel | His catch | Defense he gives |
|---|---|---|---|---|
| 1 | Nmap (network scanner) | Every device and open port on a network | ”Loud”: shows up in security logs; finds doors but can’t open them | Know your own network’s open ports |
| 2 | Wireshark (packet capture) | Any unencrypted traffic as readable text | Most traffic is now encrypted, so it’s mostly “gibberish” | HTTPS. Watch your own phone’s traffic once to see what it sends |
| 3 | Aircrack-ng (Wi-Fi auditing) | Weak Wi-Fi passwords | Fails against long random passwords on WPA3 | ”Go make your Wi-Fi password longer tonight” |
| 4 | Sherlock (username search) | Every account sharing one username, in about 30 s | Finds little if handles differ per site | Run it on yourself; vary handles |
| 5 | mitmproxy (HTTPS interception on your own device) | What your apps send in the background: location, device ID, installed apps | Needs a trusted certificate installed on the device | Never install a profile or certificate you didn’t ask for. “That is the entire attack right there” |
| 6 | sqlmap (SQL-injection testing) | Databases behind badly coded websites | Useless against properly built, patched sites | Test only your own sites; patch |
| 7 | AI video pipeline (110k stars; likely MoneyPrinterTurbo ^[inferred]) | One topic becomes a finished short | Paid API calls per video; generic output | Recognise filler content in your feed |
| 8 | Chatterbox (Resemble AI voice cloning) | Anyone’s voice from about 10 s of audio, 23 languages | Inaudible watermark that survives compression; detectable | Drop voice authentication |
| 9 | AI-provenance stripper (unnamed in the captions) | Removes the metadata, provenance tags and hidden characters that label AI output | Doesn’t defeat statistical watermarks; “more famous than it is capable” | Don’t treat metadata labels as proof either way |
| 10 | AI reverse-engineering agent (in Claude Code or Cursor; likely REA ^[inferred]) | Investigates how an app or compiled file works and breaks, on its own | Beta, one developer, inherits model errors; “how you get into real trouble fast” against real systems | Use only on software you own or may legally analyse |
Why This Matters for a Marketing Team
- Voice clones of executives or clients are cheap now. Confirm unusual payment or access requests on a second channel, not by voice.
- AI-label metadata is fragile. Proving your own content is authentic, or proving that a competitor’s is synthetic, can’t rely on metadata alone.
- Consistent handles help brand discovery but also map staff accounts. Run a username search on key staff handles and lock down old, forgotten accounts.
- Mass-produced “faceless” shorts are free to make but generic, which is Sharbel’s own verdict. Distinctive, first-party content is where the advantage is.
Try It
- Change any voice-based verification you have (bank, phone carrier) to a non-voice method.
- Search your main handle with a username lookup, and close or secure the accounts you’d forgotten.
- Check your Wi-Fi uses a long random password, on WPA3 where your router supports it.
- Make a team rule: nobody installs a configuration profile or root certificate because an app, link or person asks them to.
Open Questions
- The captions don’t name tools 7, 9 and 10. Tool 10 matches REA’s description, and tool 7’s 110k stars and “money printer” name suggest MoneyPrinterTurbo. Both identifications are mine.
- Sharbel’s demos (cloning his own voice, intercepting his own phone’s traffic) are shown on screen. The captions only describe them.
Related
- Reverse-Engineering Software With Coding Agents — REA and the legal-risk discussion
- Zero Trust for AI Agents — defensive posture when agents act on systems
- Open-Source Replacements for Ten Subscriptions (Sharbel) — the companion video’s lighter list
- AI Voice — voice models, where cloning sits