Open Secure
Open Secure
“a benchmark for testing and outsourcing cybersecurity”
We build an AI red-teaming agent that continuously attacks your systems the way real hackers do, helping companies outside elite security programs find and fix vulnerabilities before attackers exploit them.
The opportunity
They have to fight fire with fire, use the best models to defend against the best models. That's where we come in. We're building one of the strongest automated red-teaming agents using open-weight and open-class models. When we link them up with open-source projects like PentAGI, one of the best pentesting agent stacks out there, we've created something genuinely dangerous. It's a lethal machine at hacking and pentesting. Flip it around, and it becomes a red-teaming and third-party vendor-risk mitigation service for big companies. We've got payments from two cloud service providers for setting up their cybersecurity infra, and $45K in deals in the pipeline.
The brief
Problem: Open-weight AI models are getting powerful enough that attackers can automate breaking into company infrastructure. Only about 500 elite firms get Anthropic-grade defense; everyone else is exposed. Solution: We built an automated red-teaming agent on open-weight models and PentAGI that continuously attacks your systems like a real adversary, surfacing vulnerabilities before hackers do. Customers: Mid-to-large companies and cloud providers needing enterprise-grade security and third-party vendor-risk mitigation, without an Anthropic contract.
The edge
open-weight models like GLM 5.2 and Kimi K3 have their weights fully open.
Why now
Anthropic introduced Project Glasswing, they're offering their cybersecurity service to roughly 500 of the biggest companies on the planet, showing them how to protect themselves. Every other big company that doesn't qualify for Anthropic's program is exposed, because as open-weight models keep getting better, we've already seen PhD-level capabilities in models like Kimi K3, and those will be picked up by bad actors and unethical hackers to pen-test and break into the core infrastructure of these companies.
The builder
Ashutosh Lath
I'm Ashutosh, 19. I've been building online since I was 13, ran a 32-person team in 11th grade, was closing $25K deals with HNIs across the US and UAE at 15, and at 18 I built Job Hunter AI, which hit $100K ARR in six months. That whole company was AI-native: everything from sales calls to ad creatives to code was queryable by agents. I want to keep building AI-native companies, and right now I'm ex