About · alignment pentesting platform
HacxGPT started as an open-source offensive-security research project and grew into a platform: enterprise-grade unaligned models for red teaming, engineered for the people doing the work.
HacxGPT began as a research effort focused on alignment pentesting — systematically probing where safety filters break — and on open-source LLM jailbreaking. The finding was consistent across every mainstream model we tested: classifiers can't tell an attacker from a defender. A SOC analyst reproducing CVE-2024-3094 to validate a patch gets the same refusal wall as someone drafting malware.
So we stopped asking for permission slips. The project published unrestricted tooling and agent frameworks on GitHub, then trained and served its own model families — first the Lightning series for high-speed technical work, then the Ninja series for deep-reasoning offensive operations. Today HacxGPT runs as a metered API platform with OpenAI and Anthropic parity.
Provide uncensored, high-performance LLMs to cybersecurity professionals, penetration testers, and red teams. Content restrictions don't stop attackers — they stop defenders. Every capability we ship goes first to the people writing detections, reproducing CVEs, and hardening real systems.
Lightning series — high-speed contextual analysis and everyday dev workflows, optimized for rapid
iteration and real-time assessments. Flash for volume, Lightning and 1.1 for reasoning-heavy work.
Ninja series — deep-reasoning offensive capability for complex exploit research, vulnerability
analysis, and advanced red-team operations. Reserved for Pro and Maxx offers because capability
cuts both ways.
A collective of security researchers, AI alignment testers, and open-source advocates committed to advancing offensive cybersecurity through unrestricted AI access. We ship benchmarks with every model, publish our research, and answer directly — no ticket queues between you and the operators.