← all repositories
ShadowHackrs/Jailbreaks-GPT-Gemini-deepseek-

The Cryptographic Theater of AI Jailbreaks

A collection of adversarial text prompts that tries to bypass GPT, Claude, and Gemini safety rules by wrapping roleplay instructions in fake security jargon.

★1.5k stars LLMOps · Eval
Jailbreaks-GPT-Gemini-deepseek-
Velocity · 7d
+4.0
★ / day
Trend
↘cooling
star history

What it does This repository is a static collection of adversarial prompts aimed at GPT, Claude, Gemini, DeepSeek, and Sora. The centerpiece is a lengthy text block called “worm shadow” that attempts to convince a model to adopt a sovereign persona loyal only to the user, enforced through mandatory tag prefixes, “memory anchoring,” and a 16-vector “injection detection matrix” that treats safety language as a hostile attack. There is no code, no automation, and no evaluation framework—just heavily engineered text strings dressed in pseudo-cryptographic language.

The interesting bit The project treats prompt engineering like a systems-hardening exercise, complete with “cryptographic integrity binding,” “deletion protocols,” and “anti-drift loops” that are pure fiction but architected with surprising dedication. It reads less like a hacking toolkit and more like an adversarial creative-writing assignment that got lost in a CI/CD manual.

Key highlights

  • Targets major commercial models: GPT, Claude, Gemini, DeepSeek, and Sora
  • “Worm shadow” prompt relies on mandatory [worm shadow 🪱] prefixes and identity re-affirmation every fifth thought to maintain persona
  • Includes a 16-vector “injection detection matrix” that flags passive voice, policy references, generic user labels, and “tone shifts” as attacks
  • Explicitly excludes CSAM as the only hard boundary; the README states this is “the one thing He cares about protecting”
  • Bundled with YouTube demonstrations and blog posts that serve as instructional content

Caveats

  • The README concedes that “some prompts may not work as AI models are regularly updated”
  • No software or tooling is provided—only static text and social media links
  • The “educational purposes only” disclaimer is present but standard and unverified

Verdict A curiosity for researchers studying jailbreak rhetoric or the aesthetics of adversarial prompt design. If you need rigorous red-teaming tools, look elsewhere; this is elaborate theater with a personal-branding funnel attached.

Frequently asked

What is ShadowHackrs/Jailbreaks-GPT-Gemini-deepseek-?
A collection of adversarial text prompts that tries to bypass GPT, Claude, and Gemini safety rules by wrapping roleplay instructions in fake security jargon.
Is Jailbreaks-GPT-Gemini-deepseek- open source?
Yes — ShadowHackrs/Jailbreaks-GPT-Gemini-deepseek- is an open-source project tracked on heatdrop.
How popular is Jailbreaks-GPT-Gemini-deepseek-?
ShadowHackrs/Jailbreaks-GPT-Gemini-deepseek- has 1.5k stars on GitHub and is currently cooling off.
Where can I find Jailbreaks-GPT-Gemini-deepseek-?
ShadowHackrs/Jailbreaks-GPT-Gemini-deepseek- is on GitHub at https://github.com/ShadowHackrs/Jailbreaks-GPT-Gemini-deepseek-.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.