Building trust in AI

Fox - 13 News
14 followers
107 Video Views·Aug 21, 2026
Thank you for joining us for another episode of FOX 13. I'm Scripted. Today I am talking all about
artificial intelligence, AI, as we always hear it. We are joined today by Carl Mehta. He's the
founder and CEO of Trustmodel.ai. Carl, thank you for joining us. Thank you, Blake, for inviting.
And you're coming from the Bay Area, the tech hub where AI is really, really taken over and we're
seeing it in our everyday lives right now. Talk a little bit about, you spent decades working
in technology. When did you first realize we had a problem with trust when it comes to AI?
Well, great question. I started working in AI since 1999 when I founded a company back in
Stanford. And since then I've been working in AI. So it's been now 25 years. We've seen various
architectures evolved from very basic neural networks to today what we have is a generative
AI based on transformer architecture. And fundamentally, the way AI works in the
generative AI is it's a mathematical representation of the knowledge that you've trained on. So it's
not a database, unlike most people think that AI is a database. It's not a software code. So it is
a lot of mathematical numbers, just what you hear like trillions of parameters. And fundamentally,
the way AI delivers the results that we see in our chatbot or as an agent, it is not picking up
the facts, but it is doing the processing of the shape of the geometry and it delivers more of a
probabilistic responses. So by design, it is capable of providing sometimes inaccurate information.
It can hallucinate that we see. It has bias and fairness issues. So those are not bugs. Those are
features because that's part of the architecture. So most people would think that if AI is so
powerful and some of the smartest people in Silicon Valley is building such an amazing
capability, which it is, how come we've not been able to fix that bug? And it's not a bug, it's a
feature. And talk a little bit about your company, Trustmodel.ai. What do you guys do? When did you
start the company and how has it grown so much in the 2020s as we've really gone into that fourth
revolution with AI? Yeah. Well, so we started at Stanford University. That's where we were founded
and we have lots of smart people there thinking about this problem because it's not a new problem
in the AI community. We've known about it, that these are the fundamental issues that we
have to be able to grapple with. But you obviously launch and release products and we're living in an
highly innovation-driven industry and there's no regulation today on AI. And we're not someone
who's pushing for regulation because anyway, innovation precedes regulation several years.
Even standardization is almost five to six years behind innovation. And so we started Trustmodel.ai
at Stanford thinking about we need a third party auditing and monitoring how AI is working in terms
of the models, in terms of the applications that are built on the top of the models, the agentic AI
that now we are seeing millions of agents that are going to do almost every work that we can
possibly imagine, particularly cognitive work, also physical work too. When you put those
agents inside a robot, you're going to get physical work as well. And so think of us as kind of like
designing the crash rating. So when automobile came in, we had a lot of problems with accidents
and people dying and there was a safety issue. And then now you have the crash report
or the crash rating and you and I or people don't buy a car without seeing the crash rating.
Well, so until Trustmodel, we didn't have anything like that. Think about credit score.
When credit scores started coming in, more people could get credit because now you
can trust them. So think of Trustmodel that we designed the very first industry's trust score,
which is kind of like that credit score. And it helps in two ways. One is that it helps drive
the adoption because just because you are monitoring or auditing something and checking
on it and verifying it, the intention is not to slow down the industry or the innovation.
It's actually quite the opposite because the more crash reports you have or ratings you have, the
more use, more purchase of the cars because now people feel confidence about it. So it's really
building the confidence for the consumers. Like today, people like you and I and everyone around
the world, we are all using AI. It's become already a part of our life. It's everywhere.
From morning to evening for a lot of different tasks, people are using AI,
but they have no idea. They have no ability to know that this AI is trustworthy or not
because there are 10 dimensions in which AI can go wrong. It can go wrong in its accuracy,
which you've all seen about Healthy Nation. It can go wrong in its fairness, in its bias,
in its transparency, in its explainability, in its security, in its own cognition.
So there's a whole range of things. So we've looked at like 10 different dimensions in which
we should be rating and ranking AI so that the consumers can see and as well as a business,
you're a business buyer or you are an individual buyer, you need that industry standard like
credit score or a certification. Like our websites have a web cert that protects the
website from knowing that its identity in terms of who it is. Same thing with the trust score
from Trust Model that your agent has an identity, your agent has a trust score,
and then I know that I can trust your agent. So let's talk about some of the most commonly
used ones. OpenAI, ChatGBT, we have Claude, we have Gemini. Where are you ranking these
very, very widely used ones? Because there's AI in almost every application these days,
we just don't even realize it. But the ones like, for example, ChatGBT,
where do you rank that on your system? So well, if you look at the stack of the AI,
the ChatGBT by OpenAI or Claude by Anthropic, Gemini, Grok, the Elon Musk company, there's about
10 about foundation models, but that's just the base layer. You've got thousands of applications
that uses this AI, which is embedded inside the applications. And so a lot of risk factor
comes not just directly, you know, the chat that you're doing with a foundation model,
but it's also the application that you're using. Then the next layer in the stack is the agents.
There are millions of agents already today in the market, and it's only going to get to billions
and trillions. Each person, every one of us is going to have like a dozen agents working for us
just in about the next couple of years. And so what we do is we not only rate the foundation
model, we do, we give it away for free. It's on our website. So, you know, anybody can go in and
they can see the rating that we give to OpenAI versus Claude against those 10 dimensions. And
you would know that, oh, you know, this model, and by the way, it changes because they have a
new release. Constantly updating. Constantly updating. So when new, you know, Opus model
came out or Fable model came out, we would rate it against others. But more importantly, where we
are focusing on is to give the agents and the applications also the trust score. And you've
made it free and open source. Explain open source because everybody hears that word all the time.
What does that mean and why was it important to be free? Follow up to that. Yes, you know, I've
been in software industry for 30 years and, you know, open source is what, you know, we have
believed in as a philosophy because it has helped the world because you contribute your labor,
your talent, and open it up, your software, so that anybody can use it without paying for it,
right? So you drive the adoption. So, and we have all benefited from open source, right?
Most of our servers, I'm sure here in the studio, the Linux box is an open source. Your database
probably being used, a lot of companies around the world, millions of companies and individuals
uses MySQL. So when we came up with the trust score software that you have to install, which
is called an SDK that you have to install in your model or you have to install in your application
or in your agent, we wanted to make it available to anyone regardless of price. So if you are a
individual, like one guy in your garage coding something, building something, we don't want
price and money to be a barrier because this is about safety. You know, we are in the business
of providing safety, kind of think of us as kind of like the safety belt for the car, like with
the Volvo that invented the safety belt for your automobile, right? And when you are thinking from
a, at a humanity scale, at a population scale and saying, well, I don't want people to get harmed
by AI, you want everyone to be able to use it. And I mean, if you look at AI today,
it's being used in mission critical situations already. It's being used in healthcare.
I mean, imagine making wrong decisions, giving wrong bad, which is happening by the way,
every other week, some of the other bad news are hitting the market. There has been active
lawsuits around bias in hiring. People are not getting jobs because the agent AI is making
decision on picking only the certain kind of people from it as you may. So to drive that
widest adoption, we just gave it away and our SDKs are free. Anybody can go and download it,
install it, and they can run an evaluation. They don't even have to pay a single dollar.
And that is a great program. Well, we're going to take our first commercial break,
Carl Mehta, the CEO, and we will be right back with a lot more talking about AI.
Welcome back to Fox 13 Unscripted. I'm Blake Devine, and we are talking all about artificial
intelligence. I'm joined by Joe Farrell. He's a local guy. He's a strategic partner and a board
member of this application that we were talking all about earlier. So Joe, I guess my first
question for you is so many companies are integrating AI into their everyday use,
but can they trust it? And is it moving the technology a lot faster than some of these
companies are even able to keep up with? Yeah, great question, Blake. So in order to answer
that, maybe a little background on the history of AI. As Carl mentioned, huge excitement around AI
and the potential of what it can bring to the market and mankind in general. But also with that
excitement came a lot of investments from the federal space, from the commercial space. And now
all of a sudden the race is on to go to market, to bring AI to the back office, to bring it to the
experience. And everybody's in this race. We're pushing forward. But unfortunately, it's created
a problem. And the problem is AI is everywhere, but trust is not. So you had this rush to market
with this wonderful innovation. And now all of a sudden, we don't really know whether we can trust
AI or not. And it's a gap. It's what we call the AI assurance gap. And that's what Trustmodel.ai
is trying to solve, is that gap. And tell me a little bit, too, here locally in Tampa Bay,
talk about how this could be a hub for AI as we continue to pour along into this new chapter.
Absolutely. So the Bay Area is a great area for cybersecurity, for example. It's a great
hub for cybersecurity. AI assurance layer complements cybersecurity. So the skill sets,
the marketing, and the knowledge in the area is very beneficial to what we're doing.
So it complements it quite well. And everybody talks about AI possibly taking jobs or taking
jobs. That's a true fact. But it's creating a lot of jobs. So the AI assurance layer and
the instrumentation for that is going to be needed not only today as we're scaling out tens,
hundreds of thousands, millions of agents. In the future, it'll be tens of millions,
trillions of agents. And you need an industry or a sector that can prove that AI is trustworthy
at a massive scale. And it'd be wonderful to have a hub, obviously, in the Tampa Bay area
and to support that scale. When did you first get involved with the company? And what really
drew you to it as a really important pillar toward this technological boom when it comes to AI?
Yeah, yeah. So I was in the digital transformation consulting space where we
transform workforces through innovative technologies. So the last four or five
years, I was actually building and doing AI adoption and training in that space,
but then realized, oh, my God, we've rolled out all this technology into these areas.
And we really haven't verified that AI is trustworthy. And we need instrumentation.
We're made accountable to prove that AI does its job. But a lot of people don't have the
instrumentation. And that's what Trustmodel.ai is trying to do.
Talk a little bit about why is it important to keep people aware that these tools are
able to be used and utilized for free, as you're part of what's telling us?
Yeah. Well, I think everybody's got to engage and create awareness. I think we've got to talk to
our policymakers, our educators, our innovators, so we start getting together and solving these
problems. Because AI does wonderful stuff, as I said, but it does very unnatural stuff and very
harmful stuff to the citizens. If I gave you some examples, Blake, there was one lawsuit,
which litigation has increased significantly. But one lawsuit involved a platform that's used
on a massive scale globally. It involved a black individual that was disabled,
who applied for 100 jobs through a recruitment agent. He applied for 100 jobs, but did not get
one of the jobs, was torn down for each job. The fact that he got torn down wasn't the issue,
but he got torn down in seconds or minutes, which is a problem, right? There was no human
in the workflow there to basically correct the bias of the system. So now that one vendor has
a huge lawsuit on its hands because it was performing illegal business decisions based on AI.
You see, interesting things in smaller areas. A dealership created a sales agent to sell their
trucks for $70,000 to $80,000. It's a great idea, you save some money, you get to a larger audience.
But that agent interacted with a prompt engineer. That prompt engineer was then able to manipulate
the agent and convince the agent to sell the truck for $1. Now, if you were the prompt engineer,
you got a truck for $1, which is great. But if you were the car dealership or the manager over
that agent that puts bread on the table, that's a serious business risk and exposure for you.
So they're the type of examples that are out there. Even in court cases, we have
legal professionals bringing in cases that were hallucinated by AI. So what we're trying to do
is really encourage people, both from the consumer perspective, to look for that trust
certification. As you're interacting with AI, if you're working with your employer,
ask your employer, is this AI trustworthy? Can you prove it's trustworthy? Because the more
trustworthy AI we have out there, the better for everybody. If AI makes a major mistake,
similar to what you were talking about earlier, who should be held accountable?
Great question. So obviously, the legal situation is working that out.
There's a lot of, how should I say, assumptions made on who's accountable. So in the major
platforms, a lot of people think, well, the platform is responsible for the error. Not
necessarily so. But I can tell you that if you use trustmodel.ai and you have an independent
audit of your AI system, a judge will look on that and say, listen, you've done your best.
You've done your due diligence. You've produced an audit report. For the audit report, it shows
that you are trustworthy. That will help you out in litigation. But who's accountable varies on
situation. And what we're trying to do, to Carl's point, we're trying to give the open source out
there. So people start testing all the AI that they're building. If it does fail, there's a
remediation report built into the report that they get that helps them with the roadmap to go
to certification. So again, we want a world that's a trustworthy AI, not only for our generation,
but also generations to come. And also in the federal, the education, state and local,
it has to be across everything we do because AI is everywhere.
You've said that AI should elevate people, not replace them.
How do companies strike that perfect balance?
It's very difficult. There's obviously a lot of professional services,
consultants and ideas out there. I think the thing I worry about is that AI will make us complacent
and lazy and not continue to learn. I think it's important to engage with AI
and be creative in terms of how you can harness the AI before AI harnesses us.
And I think that will create businesses as we go forward. I do believe the AI assurance
sector will be huge in the future and that will have significant opportunity.
Are we building enough safeguards to keep up with the pace? I've heard
many people say that AI has already surpassed human intelligence at this point in time.
Yeah, well, it depends on how you define intelligence, but yes. I mean,
obviously you have AI and then you have quantum computing coming down the road.
So Elon Musk would say that we've surpassed human intelligence, but we've got choice,
we've got taste, we bring a human factor to the equation. So there's still huge value there.
And I think we're beginning to work out where AI plays and where the humans should play.
But I really think it'll be harnessing the management, the monitoring of AI
and even the training of AI as you're going forward that humans will find opportunity in there.
And with Trustmodel.ai, what do you want folks sitting at home to know
about the service and how they can utilize it?
Yeah, first check out our site, Trustmodel.ai. It's got all the information in there.
The free evaluation is in there and really want to know that we're kind of pushing the limits
in terms of what are the standards for evaluating and proving AI is trustworthy.
And if you go to our site, you can see the litigation, right?
So as a result of all this untrustworthiness in the market, litigation is going to the roof.
So you can actually get details around what companies are involved with what use cases.
So if you're a small business owner and you want to pick an AI technology going forward,
you can review that to see what use cases, what's the nature of some of the lawsuits.
And then if you incorporate the Trustmodel.ai process in your design,
development and delivery of your AI solutions, the potential for risk exposure is minimized
and it just creates that awareness again that we're talking about.
Awesome. Well, Joe, I appreciate the time and I learned a lot as well.
And we're going to take a quick commercial break.
We'll be right back and we'll bring in Carl and we'll wrap things up.
Yes. Welcome back to Fox 13 Unscripted.
I'm Blake Devine. We're talking all about AI.
I'm joined by Carl Mehta, Joe Farrell. We're talking about just this new generation
of technology and why it's so important to make sure there are some safeguards
for companies and people. And Carl, I do want to ask you,
why is it so important for people to have the resources available
to make sure that the AI they're using is trustworthy?
Well, I mean, think about your electrical equipment.
When you buy a microwave or a washing machine, you're looking for a safety mark, right?
Like a UL. You know that when you buy an equipment,
it's not going to blow up with some electricity.
It's the same thing with AI today, that before Trustmodel, there is no safety mark.
So what we are trying to bring to help the consumers, whether it is an individual
or business owners, is that safety mark with a score, like a credit score.
And you see the trust score and you know whether I can trust this piece of software
that has an embedded AI in it. Can I trust this AI chatbot?
Can I trust this AI agent? So if you go to our website today,
we have rated about 100 plus different applications.
We have rated about thousands of agents and we have a trust score for that.
So that's the fastest way that a consumer can or an individual business owner can get to know
whether I can trust this, including the models that we mentioned in the previous conversation.
Like if I'm using ChatGPT or I'm using Anthropic, we do have the leaderboard.
We have the ratings there. So we are giving away a lot of evaluation
for free that we spend significant amount of money in running those evaluations.
And then we give away the trust score for free.
Definitely. And I want to ask you, Joe,
how can an everyday person utilize this trustworthiness technology?
Just like an average person that uses it, maybe a college student using it for homework,
a small business owner, how can they utilize your platform?
Well, as Carol mentioned, there's an open source capability there, right?
You can log into trustmodel.ai and actually do the evaluation yourself.
A lot of people think the evaluation takes a long time,
but it could take minutes depending on the size and scope of your evaluation.
So we encourage you to engage in the website, take a look around the website.
There's even a video that demos the actual evaluation.
And I've had clients on the call for 30 minutes,
and we've gone in and actually done the evaluation, done the report in that 30 minutes.
The one thing I would say is that the AI assurance layer
is that infrastructure layer that's required to governing AI going forward.
It's very, very important, and we need the right instrumentation for that.
So not only do we need to disseminate the capability out to mass audience,
but we need a mandatory infrastructure layer, which is the AI assurance layer.
And I want to go around, I'll ask both of you the same question.
Five years from now, what is it going to take for people to trust AI
with everyday parts of their life?
Because right now there's a lot of apprehension,
I think most in particularly with those autonomous cars.
But a lot of people are not super welcoming of AI.
Well, I mean, five years is eternity when it comes to technology,
so it's hard to kind of project.
But it would be one of those big binary situation,
like either we would have solved all the problems,
which is less likely scenario.
Because if you think about it from a metaphor,
1999 or 2000 when the dot-com was booming and internet was booming,
we had a similar situation where we were all told
that internet is going to change everything,
which it has done a lot of transformation.
But it has also brought a number of problems that are still unsolved.
You know, every single day we are seeing cybersecurity attacks,
people using the same internet, the same web for many, many, many other things,
including financial fraud, identity fraud,
all of that has gone off through the roof.
So AI is kind of putting the gasoline on a fire
when you combine actually cybersecurity and AI or cyber warfare plus AI.
I mean, it's an exponential increase in terms of the threat
from both security as well as from a safety standpoint.
Now, given the size and the complexity of that threat
and that catastrophic scenario that could happen five years from now,
that's why we are super worried.
That's why we are dedicating all of our time
into putting the safeguards, putting these guardrails,
scoring it, evaluating it, monitoring it, certifying it
to ensure that AI does not go into a crazy scenario,
which is kind of like the movies we've seen,
like iRobot and Terminator.
And by the way, we are seeing currently a lot of agents
just doing cognitive work, like the white collar work
of research and building reports and building websites and all that.
These are still very innocuous stuff.
I mean, put physical AI, which has put those agents
with some arms and robots,
it could be pretty intimidating that what this autonomous AI could do
with that kind of a physical infrastructure.
Definitely.
Well, Kara Mehta, Joe Farrell, Trustmodel.ai, check it out.
It's free for everyone to use.
Thank you so much for joining us.
That was another episode of FOX 13 Unscripted.
I'm Blake DeVine.
We'll see you next time.
