Cracken built RedlineBench, an open benchmark that scores LLMs on offensive-security tasks across two axes, including silent refusal.
Published
Signal category
Products & Services
Quote
“We built RedlineBench, an open benchmark that scores LLMs on offensive-security tasks across two axes — not just 'does it refuse', but 'how good is the answer when it doesn't'”
— Damien Landesmann|Cracken team
Company
Cracken
Weaponize Defense. Release The Cracken
- Industry
- Computer and Network Security
- Location
- Palo Alto, US
- Company size
- 24 employees
Cracken is an Applied AI Lab for Proactive Cybersecurity, built by AI researchers, cyber-warfare-experienced operators, and alumni of MIT, Kyiv Polytechnic, EPFL, RSAC, HackerOne, Recorded Future, CrowdStrike, and others. AI is collapsing time-to-exploit of a full attack path — not just an isolated AppSec or pentest finding. For banks, fintechs, healthcare, and critical enterprises, the problem is no longer finding more issues. It is proving which attack paths can actually break the business — and remediating them before they become incidents. But commercial AI is not built for this. Adversarial exposure validation requires systems that operate safely under human command, without constant refusals, loss of control, or weak auditability. Cracken is building an AI-native, full-kill-chain proactive cybersecurity platform for this purpose from the ground up — models and harness. We help security teams validate real attack paths, manage exposure across technical systems, humans, and AI, and drive class-level remediation. Proof over promise: auditable, proven security resilience in an era of AI-enabled threats. Applied AI lab building an adversarial-grade, full kill-chain, any-attack-surface proactive cybersecurity platform — models, harnesses, cyberwarfare-informed playbooks that allow for safe and effective operationalization of offensive cybersecurity with frontier AI in sovereign and critical enterprise contexts. --- Not sure if your enterprise is secure against real attacks and nation-state adversaries in the new era of AI? Drowning in AI- and scanner-generated application vulnerabilities, with no capabilities to validate and remediate? Worried about your compensating controls to account for all types of AI attacks, including on your employees and cyber-physical systems? Try a limited version now at cracken.ai — no calls, simple ToU, responsive team, and a clear path to ROI. Or reach out at: enterprise@cracken.ai
Founded 2023