EXHIBIT A: THE GUARDRAIL PEOPLE
Company Profile • Enterprise AI Security
Dynamo AI
The safety inspectors for enterprise artificial intelligence - the ones who show up before the model does something it cannot take back.
The safety inspectors for enterprise artificial intelligence - the ones who show up before the model does something it cannot take back.
Somewhere right now, a compliance officer at a Fortune 500 bank is staring at a generative-AI demo that works beautifully ninety-nine times out of a hundred. It is the hundredth that keeps her up at night - the answer that leaks a customer's data, invents a number, or cheerfully follows a stranger's instruction to ignore all previous instructions. Her job is to say no until someone can prove it is safe. Dynamo AI is the company that hands her the proof.
Dynamo AI sells the unglamorous layer that sits between a powerful model and a regulated business: a system that tests AI for the ways it can fail, wraps it in real-time guardrails, and produces the paperwork auditors and regulators ask for. Its customers are not hobbyists. They are banks, insurers, chipmakers, and the U.S. Army - organizations for whom a hallucinating model is not a funny screenshot but a headline.
The large language model is a remarkable invention with one inconvenient habit: it will say almost anything, fluently, whether or not it should. It can be talked out of its own rules. It can repeat private data it was never meant to remember. It can fabricate a citation with the same steady tone it uses for the truth. For a consumer app, that is a quirk. For a bank, an insurer, or a government agency, it is a liability with a dollar figure attached.
And the dollar figure is rising. The EU AI Act, with enforcement of its prohibited-practices provisions beginning in February 2025, threatens penalties of up to 35 million euros or 7% of global revenue. Regulators in other jurisdictions are sharpening similar pens. The result is a standoff: the business side wants the productivity, the risk side wants a guarantee, and the technology, left to its own devices, offers neither.
This is the tension Dynamo AI exists inside. Not "can we build AI?" - that question was answered years ago. The question now is "can we trust it in a room full of lawyers?" Most of the industry treated safety as a feature to bolt on later. Dynamo AI treated it as the product.
Dynamo AI - originally DynamoFL, where the FL stood for federated learning - grew directly out of Vaikkunth Mugunthan's doctoral research at MIT's Computer Science and Artificial Intelligence Laboratory. His thesis subject was privacy, security, and compliance in AI systems: not the part of machine learning that gets the magazine covers, but the part that decides whether the technology is allowed indoors. He co-founded the company in 2021 with Christian Lau, then also a PhD student at MIT.
The bet was contrarian for its moment. In 2021, the excitement was all about making models bigger and more capable. Mugunthan and Lau wagered that the bottleneck would not be capability but trust - that the companies with the most to gain from AI were precisely the ones forbidden from using it carelessly, and that they would pay for a way to say yes. It was a bet that compliance, of all things, would become a growth market.
Investors agreed. A $4.2M seed round was followed in August 2023 by a $15.1M Series A led by Nexus Venture Partners and Canapi Ventures, bringing total funding to roughly $19.3M. The angel list read like a who's-who of people who worry about data for a living: Apple's lead for privacy-preserving machine learning, Dropbox's head of governance and risk, a product leader from Snowflake. When the privacy professionals are writing personal checks, the thesis is doing something right.
Vaikkunth Mugunthan and Christian Lau spin the company out of MIT CSAIL research on privacy-preserving machine learning.
Early backing to build privacy penetration testing and compliance tooling for enterprise LLMs.
Led by Nexus Venture Partners and Canapi Ventures, with Formus Capital and Soma Capital. Total raised reaches ~$19.3M.
The platform broadens into full AI security, evaluation, and governance. Featured by Comcast NBCUniversal LIFT Labs.
Launches out-of-the-box guardrail models aligned to Article 5's prohibited AI practices, ahead of enforcement.
Dynamo AI's platform is best understood as a pipeline of skepticism. First you find out how the model breaks. Then you stop it from breaking in production. Then you watch it, forever, because models are not famous for staying the same.
A testing and red-teaming suite that probes AI systems across 20+ vulnerability classes - the structured way of finding out what your model will do before a stranger does.
Customizable, real-time guardrails that catch compliance violations, data leakage, hallucinations, and jailbreaks - reportedly detecting unsafe prompts 2-3x better than Llama Guard.
Risk controls for agentic AI. Autonomous agents take actions, not just answers - so they need a chaperone with automated risk detection.
The clever part is the customization. Off-the-shelf safety filters are blunt; a bank's definition of "unsafe" is not a hospital's, and neither matches an army's. Dynamo AI starts with deep policy research - reading the actual regulation and the actual customer's rules - and turns that into guardrails that can run out-of-the-box or be tuned, even on-device. The result is governance that fits the business, rather than a business contorting itself to fit a generic filter.
The strongest evidence for a safety company is the caliber of the people willing to trust it with their reputations. Dynamo AI's named roster skews toward organizations that do not buy on hype.
These are the financial services, insurance, consulting, consumer electronics, and automotive customers - many of them Fortune 500 - for whom an AI mistake is a regulatory event, not a tweet. The company holds ISO and SOC 2 certifications, and a co-founder, Dr. Christian Lau, has testified before the U.S. Congress on AI policy. When you are advising the people who write the rules, you tend to understand the rules.
A 49-person company standing between trillion-dollar enterprises and a 35-million-euro mistake. The leverage is the whole story.
Dynamo AI's stated aim is to let organizations productionize generative and agentic AI with confidence, through auditable guardrails and risk management. Strip away the phrasing and it is a simple promise: turn "we cannot use this" into "we can use this, and here is the evidence." The company's bet is that as AI regulation tightens around the world, the demand for that translation only grows. Stricter rules are not a headwind for Dynamo AI. They are the business plan.
There is a quiet ambition underneath. If the safe path and the fast path become the same path, AI stops being a thing companies fear and starts being a thing they simply use. That is the world Dynamo AI is building toward - one where the compliance officer's hundredth case is handled before she has to ask.
Today's debate is about chatbots that talk. Tomorrow's is about agents that act - software that books, buys, sends, and decides on its own. The blast radius of a mistake grows accordingly, which is why Dynamo AI built AgentWarden before most enterprises had agents to warden. The company is positioning for a future where AI is not a feature you demo but an employee you supervise.
Return to that compliance officer at the bank. A few years ago she said no, and she was right to. Now she has a system that tests the model, guards it in real time, and hands her an audit trail. The demo that kept her up at night is in production. The hundredth case - the one that leaks, lies, or obeys a stranger - is caught before it reaches a customer. She still does not fully trust the model. She does not have to. That is the entire point of Dynamo AI.