“Firewalls Won’t Save You from ChatGPT”: Meet the Startup Building the First Security Layer for GenAI!
AI is no longer in the labs. It’s in your workplace. Your mobile apps. Your kid’s school project. But here’s the real problem:
Now imagine: What if someone could poison your AI tool’s training data, steal sensitive prompts, or manipulate answers? That’s not sci-fi. It’s already happening. That’s exactly where Repello AI comes in. The San Francisco and Bengaluru-based startup just raised $1.2 million in seed funding to build the world’s first “always-on” security system for GenAI.
Let’s break down what they’re doing, why it matters, and how they plan to lead this next frontier.
Who Just Backed Repello AI?
The funding round was led by:
- Venture Highway (acquired by General Catalyst)
- pi Ventures
- Entrepreneur First
- And angel investors like:
- Charles Songhurst (former Microsoft exec, Meta board)
- Vivek Raghavan (CEO, Sarvam.ai)
- Satya Vyas (CEO, Project Hero)
That’s not your average seed check. These are some of the sharpest minds in AI and deep tech backing this bet.
The Brains Behind Repello AI
Founded in 2024 by Aryaman Behera and Naman Mishra, both IIT Roorkee alumni, Repello AI isn’t just another “AI for AI” startup. Their goal? To proactively protect GenAI systems before they get hacked, poisoned, or manipulated.
Naman (CTO) adds:
So What Exactly Are They Building?
Let’s get a bit nerdy (but keep it simple). Repello AI currently has two core products:
1. ARTEMIS
Full Form: Automated Red Teaming Engine for Mapping, Identification, and Scanning.
Sounds fancy? Think of it like an army of ethical hackers who are constantly poking, prodding, and stress-testing your AI systems, but automatically.
Key Features:
- Runs millions of adversarial tests on your AI models
- Works across text, image, and audio
- Finds vulnerabilities before hackers do
- Delivers actionable fixes and compliance reports
Why this matters: Traditional security is reactive. ARTEMIS is proactive. It doesn’t wait for the fire; it finds the gas leak before the match is lit.
2. Repello Guard
If ARTEMIS is the watchdog, Repello Guard is the real-time bodyguard. It monitors everything that the AI is taking in (inputs) and what it’s giving out (outputs).
Features:
- Detects things like:
- Unsafe or biased outputs
- System prompt leaks
- Even if your AI starts mentioning competitors (yup, that’s a real threat now)
- Can auto-filter or flag results based on customizable rules
Think of it as the “firewall + bouncer + compliance officer” rolled into one.
Wait, Are These Attacks Real?
Yes, and they’re only getting worse. Let’s look at some actual threat types:
| AI Threat Type | What Happens |
|---|---|
| Prompt Injection | Hackers trick the AI into revealing confidential info |
| Data Poisoning | Attackers sneak harmful or biased data into your training sets |
| Jailbreaks | Users force AI to bypass its own safety filters |
| Sensitive Info Leaks | AI accidentally reveals internal company data or system prompts |
| Output Manipulation | AI starts producing biased, unsafe, or misleading results |
Why The Market Needs This Now
We’re in the early innings of AI.
But here’s the big picture:
- The GenAI market is expected to cross $100 billion by 2027.
- Yet, less than 15% of AI products today follow any security standard.
- Enterprises are rushing to adopt tools like GPT, Gemini, Claude, Midjourney, but without guardrails.
- And compliance laws (like the EU’s AI Act, HIPAA, and GDPR) are catching up fast.
Repello AI is not just solving a technical problem. They’re solving a market timing problem. They’re building trust, which, in an AI-driven world, will be more valuable than code.
The Founders’ Superpower: Security + Scale
Aryaman and Naman aren’t new to this game. They’ve been dabbling in security and AI since college.
What makes them unique:
- Deep technical foundation (IIT + open source projects)
- Built AI tools and broke them, before AI even became mainstream
- Obsessed with scale + automation (their stack already runs adversarial tests at scale that most enterprise tools can’t yet match)
Plus, they’ve built their platform to be developer-first, meaning it can plug into OpenAI, Hugging Face, Google’s PaLM, or any proprietary LLM.
It’s not a tool. It’s infrastructure.
What’s Next for Repello AI?
The funding will be used to:
- Expand their red teaming and threat intelligence platform
- Grow the engineering and research team
- Strengthen go-to-market efforts
- Build integrations with cloud platforms and GenAI tools
- Form partnerships with enterprise AI adopters
We hear they’re also working with a few top Indian fintech and MedTech players for pilot deployments. And yes, they’re hiring.
Conclusion: A Startup Born for the AI Age
AI is changing everything, how we write, design, sell, and even how we think. But what systems do we use to protect it?
Still stuck in 2015.
Repello AI is not trying to patch an old system. They’re building a new one, where security scales with intelligence. In a world where a single prompt can cause millions in damage, proactive AI security isn’t a luxury. It’s a necessity. And if their early traction and investor backing are any signal, Repello AI might just be the company writing the rulebook on how to protect the AI of tomorrow.
