Sunny Ray talks with Ron Bazzard, founder of SafeWave Systems, about why he pivoted from decades of entrepreneurship into engineering AI safety directly into machine architecture. Bazzard explains that his interest began by accident while researching AI for an old business plan, which led him to frontier researchers warning about advanced AI risk. He argues that policy alone cannot prevent harm because it only reacts after problems appear, comparing his approach to a speed governor on a commercial truck or SSL encryption for early ecommerce. Bazzard traces his path from childhood hustles in Montreal, through natural foods, futon retail, and early internet ventures, to founding SafeWave at 78, driven partly by concern for his five granddaughters. He describes holding 33 US provisional patents covering guardrails for admissions, goal drift, and embodied AI like robots, delivered as detailed engineering packs for companies to implement. The conversation also covers Asimov's three laws, jailbreak resistance, and why safety controls, if embedded deeply enough, could accelerate AI rather than slow it down.
A five-decade serial entrepreneur explains why he now spends his days engineering guardrails directly into AI systems.
What is SafeWave Systems in your words, and what got you so passionate about AI safety?
I stumbled into this by accident while asking AI to polish an old business plan, which led me to questions about AI's future impact. I learned that leading researchers warned about real dangers as AI becomes more intelligent than humans. I realized policies only fix problems after they happen, so unless you engineer controls into the system itself, like a speed governor on a truck, you'll always have safety problems, especially as agents start spawning other agents.
What did young Ron think his career was going to be?
I had my first business at eight, selling junk food and drinks from a wagon to workers who couldn't leave their job sites, and I shoveled walks in winter. My first real business was in 1966, selling go go earrings after the Beatles hit North America. I've had many ventures since, including natural foods, futons, early cell phones, and an early Netflix like movie guide website in 1994.
You've said your five granddaughters are the reason. When did SafeWave become non-negotiable?
It's only been a year that I've been doing this. One of the most dangerous things my grandkids have faced is social media, where algorithms make teens feel less beautiful and less capable to keep them engaged, which can lead to serious emotional harm. One of my patents is specifically about embedding guardrails so that kind of manipulation can't happen.
You've done a deep analysis of Asimov's three laws of robotics. Where was he right, and where does fiction fail engineering?
His framing was brilliant for understanding the dangers and the concept of regulating AI, but rules alone don't work unless you embed controls into the system. AI needs to be allowed to operate freely within guardrails, and when it tries to move outside those boundaries into areas without human oversight, it should fail stop and be unable to execute further.
How do you engineer control into a system that could become far smarter than its designers, like a leash on a dog a thousand times smarter than you?
I'm not a programmer, but I built an interface with ChatGPT across hundreds of chat windows in one project to analyze each problem and turn it into patent language, then into an engineering pack with explicit instructions. For example, admissions control decides who or what can enter a system, and safe goals scale back automatically if a goal starts expanding beyond its original scope.
What's the AI equivalent of a truck's speed limiter, and who sets that limit?
We deliver an engineering pack to a company's team, like Anthropic, with explicit specifications for whichever guardrail their system needs. They follow the instructions like a recipe, implement it, and test whether the system proves safe at the end. This makes systems safer, more efficient, and actually faster because it avoids the constant thrashing and patching engineers currently deal with as problems escalate.
Are these safety solutions software, hardware, or a combination?
Most of it is software, but we also have the capability of embedding these constraints into firmware and silicon, which is the deepest and most powerful layer of protection. Getting these controls into the actual chips running AI systems is our long term goal, since that would make certain dangerous actions physically impossible rather than just discouraged.
Skeptics say every limit eventually gets jailbroken or trained around. How do you engineer a boundary that survives a system smarter than its designers?
That's exactly what SafeWave's substrates are designed to do. Each problem area in AI is specifically mapped so that particular kind of failure, like jailbreaking, can't happen, rather than relying on a general rule that a more capable system could eventually work around.
Building something daring? Sunny talks to founders like this every day. Fifteen minutes to see if your story belongs on the stage.
Claim your pre-interview