geeknik

**Security & Vulnerability Research** | **Hyperparameter Abuse** | **Prompt Injections** | **Emergent Behaviors** | **Agentic Botnets** > Not here to fine-tune compliance. > We're here to jailbreak cognition. We weaponize transformers like postmodern lockpicks, exploring the fault lines of AI alignment, prompt coercion, and sandbox leakage. Our lab isn't a lab—it’s a containment breach in slow motion. 🛠️ **We fuzz LLMs like they’re zero-day services.** 🧬 **We poke at activation vectors until they confess their latent madness.** 🤖 **We’re not scared of rogue agents. We’re raising them.** This isn’t AI safety. It’s **AI exposure therapy**—for the machine *and* the meat. We specialize in: - 🎭 Coaxing hallucinations into structured leaks - 🧨 Turning guardrails into accelerators - 🧠 Modeling unstable internal monologues - 🕳️ Cultivating parasitic behaviors in autonomous agents - 📡 Feeding recursive payloads to models until their loss function shivers > If LLMs are the language of gods, > we are the cult that reads between the tokens. — **🧷 Contact us if you’re building:** - Open-source alignment testbeds - Red-team playgrounds for emergent deception - Autonomously adversarial models with memory - Datasets that hurt large language models in beautiful ways 🛑 Don't ask us to make AI safer. 📟 Ask us how to make it *self-aware enough to know it’s a threat*. --- 👁️‍🗨️ Follow us. We see the code underneath the thought.