2026 Predictions | Episode 35

Hello, and welcome to another episode of AI security ops with Black Hills Information Security and, of course, our illustrious cohosts, miss Brumwin Acre, Derek Banks, and doctor Brian Furman, and, of course, myself, Jalf Theier. This episode was brought to you by Black Hills Information Security. If you have an AI, implementation that you're looking for some security assessment work to be done on, please feel free to reach out to us at black girls infosec dot com and fill out the contact us form, and we'll see if we can help you out. Today's episode is going to be one of these really interesting things where we try to look into the crystal ball, and we try to make predictions about what's going to happen with artificial intelligence technology in 2026. So we're gonna try to address one prediction each because I suspect this is going to turn into quite a discussion.

Alright. Yeah. I'll grab this one. So prediction, which is that in 2026, grid power not GPUs becomes the binding constraint in some regions pushing at least one major cloud to introduce energy aware AI computing price or scheduled inference discounts tied to off peak electricity. So I absolutely think that that the grid is going to become I mean, it already is becoming a major constraint.

Yeah. I was listening to a article yesterday, I think it was on on public radio, about the nuclear re renaissance that's that's happening associated with this, which is a little scary to me because there's reasons for that. I'm a very, very close family friend, has been a very anti nuke for a long time, and she's a very public figure. You know, putting that aside, what I'm actually wondering that could happen is with the advent of the development of small modular nuclear reactors, which is which is pushing forward very, very quickly, I wonder if we're gonna end up in a situation where there's gonna be some regulations that actually request the data centers power themselves instead of tapping into the grid, especially with the the small modular units coming online. So I think it's a it's a significant problem.

Yeah. From what I read in The US or in North America, we really have two options realistically. One is nuclear, like y'all just mentioned, and the other would be hydroelectric. We have a lot of hydroelectric that we could use. When you're talking about impact to the environment, I would much rather have, you know, high end modern nuclear development than I would putting a dam on every river, because I think that has a much broader, impact.

And that would be a disaster. Right? Obviously, environment.

And and and so, I mean, honestly, if, you know, if I were the one making the call for direction of, you know, US energy policy, it would be all of the above. Let's do it all and then work towards, you know, short term, long term type solution. But, you know, I don't have faith that we do things that the most efficient and best way. My wife works for the government, so,

you know

Things happen. Right? Brahmin, thoughts?

Like you, the nuclear thing is concerning, but I kind of hope and and this is maybe Pollyanna, maybe a little too much of those colored glasses, but I really do hope that the increased energy requirements of all of these AIs will stimulate better energy solutions that will turn around and provide positive dividends for humanity. I don't have a lot of faith that that's gonna happen, but you know it would be really nice and I think that having it leveraging the computational capabilities of AI to address the energy shortcomings would be a really good application and could potentially make life better for just normal people. Like I said, don't have a lot of faith,

but I would totally welcome. Kind of a corollary thought to this whole area is in 2026, do you think we're gonna see the first orbiting data center that is solar powered?

I mean, that's probably what, X is gonna do.

They're working on it.

I think that's what. Yeah. Well, I wanna say before we answer that, I I I wanted to say one more thing on nuclear. Like, I I definitely understand why people are antinuclear. I mean, you have really bad things that have happened. Chernobyl, the Fukushima plant, 3 Mile Island, for sure. Right? But I've lived in where I am in in Hampton Roads area for, what, twenty years now, which isn't a secret. I don't care if people know where I live. You can find it out if you really wanted to.

Yeah. I I would if it's possible, an orbiting data center is a fascinating idea, but I I think there's so many logistical problems to solve to get us there.

So well yeah. But, you know, life life imitates sci fi. Right? I I I I think I mentioned in the last podcast, I read a lot, or maybe it was before we started talking. I read a book called Delta v or Delta five by Daniel Suarez.

And how would you go about shielding the materials? I mean, okay. So we're going off on our tantrum. Anyway That

that could be radical.

Just Yeah.

Okay. Any final thoughts before we move on to the next one? I think we we've covered that. Alright. Let's throw the next question to Brahmin. You wanna pick one of the predictions, or would you like to make one of your own predictions, Brahmin?

We have so many options.

And it was almost like data overload. It did such a great job. So while Robin is looking, I just wanna tell everybody that before this, I thought it would be an interesting exercise to have ChatGPT auto mode give me a prompt for research mode, and then I put that research mode prompt in the research mode, and it thought for fifteen minutes. And the output that came back was way way more detailed than I expected.

It did it did really well but I saw the prompt that you fed into that. That was a that was a very sweet prompt. So I'm I'm not surprised by the volume or quality of the output.

So get him the bumper sticker prompting hero. Yeah. Anyway, go ahead, Bramwin. Make a prediction.

So one of the predictions here is that in 2026, the FDA qualifies at least two more AI tools as drug development endpointsbiomarkers where federal data consortia expands to bypass data sharing and IP constraints. Okay. I'm not so convinced about the the the second half, but this concept of the FDA qualifying and allowing AI drug development tools and allowing big pharma to go ahead and use AI to generate and possibly even test new drugs for things. I'm already seeing a lot of stuff in the various newsletters that I follow about AI being applied to this end. And this is where the the innovations in AI that have come about in the last five years are truly exciting because the ability to potentially eliminate animal testing for different drugs, different procedures, cosmetics, Things like that are absolutely best case scenarios where AI can be applied to make life better not just for humans, but for other critters on the planet.

I think that's one of the most positive and amazing applications of the technology that we have. It's a very human centric, realistic thing to have AI assistance in solving some of these drug development and treatment development biological problems in a space where a lot of that research takes a tremendous amount of time. And one of the most fantastic benefits of artificial intelligence is it speeds up processes. Right? It's a task accelerator, and so that's a really nice development that I think can be a great benefit to humanity.

I mean, I think it's I think it's already been happening. Right. Right? I mean, you know, anybody remembers SETI at home. I mean, folding at home.

Yeah. Yeah. The fold folding at home is what came to mind to me as well. I remember it. I mean, like, way back in, like, undergrad, I remember reading about it. I honestly didn't really understand it, but it was just like, hey. If you could spare some GPU cycles for the greater good of humanity, go ahead and install this program. I was like, oh, it sounds good.

And then you're like, why is my electricity bill $40 more expensive?

Yep. Well, all your

CPUs every time you're not using Well, don't worry.

You could do, like, a $40,000 solar project to solve that problem.

Right? Oh, If I remember correctly, studying at home was mainly leveraging distributed computing, not shy. Sure.

But I I I was sorry. I was making the point that that, it's a natural progression from things that, you know, were being used that were technology wise and, you know, through machine learning and data science, you know, here recently up to using large language models. Like, I guess what I was saying is these companies have already been, you know, doing this kind of stuff Yeah. I think.

Yeah. Well, on that note, Derek, I think it's your turn to make a prediction. So

Yeah. I'm gonna off road. I I'm gonna I'm gonna pick this I'm gonna pick a cybersecurity one. And in 2026, we're gonna see nation state threat actors using AI and agents to scale patient patients' access and deniability. We already saw the first, almost autonomous, I think the 90% autonomous, use of Quad, and we're gonna see, criminals continue to scale revenue, speed and persuasion, you know, for phishing and deepfakes.

That's right.

That's That's right. I I thought there'd be a security one in the list, but I couldn't find one. So I had to go back and ask again.

I don't know I don't know that I have any comments on that. Anybody else got any thoughts on that that idea?

Yeah. I mean, absolutely. It's it's certainly already already happening. We have lots of instances that are out there that we can see of people leveraging AI for these various campaigns. As, you know, Derek mentioned, there's, you know, one of the agentic instances we've seen of a clot that Anthropic documented.

Well, on that note, I'll go ahead and make my prediction because it's a great segue. And my prediction is that during 2026, we are going to see agentic AI applications be the dominant paradigm in all application development, and the key characteristic or key feature of agentic AI implementations is going to be autonomous actions with very little human in the loop or human control involved. And I think as we go down that road of multiple multi agent systems making autonomous decisions as part of application workflows, we are going to see some applications start to exhibit a lot of really nondeterministic behavior and all kinds of security risks, as well as quality assurance, quality control entity risks show up, if you like, in trying to test these applications. And we are going to enter a world where building these animals is going to be extremely challenging, but it's going to be happening at a very rapid pace and it will become the dominant paradigm.

So what you're saying is that companies are going to use distilled internet knowledge on coding to write production applications. What could possibly be? Well,

partially I was saying that, you know, I think we've already seen in 2025 that the idea of AI assisted coding, otherwise known as vibe coding, has become a dominant thing in all application development.

Now you're saying there won't be a human vibing it, it'll just be the AI doing it doing the whole

Right. Life I think we will have giant ecosystem of AI agents out there both in proprietary applications as well as even libraries, if you like, of very public agents that people will be able to use as part of an agentic AI multi agent architecture that it's going to just explode during 2026. That's my prediction.

You know, that's actually funny that that you say that because the one I didn't pick for ChatGPT giving me a security related one was basically that threat actors were the the main source of AI compromise is going to be threat actors compromising agents and and building, you know, backdoors in through agentic type capabilities. It's like, that's that sounds scary. I don't wanna do

that one. Yeah. In the agentic space, of course, and I and I said this actually before we got on the show, from the perspective of security risk, we're going to see what I think of as a multiplicative effect of nondeterministic output, I. E, each of these agents will be able to make autonomous decisions based on their own usage of LLM technology to achieve their goal or objective, but you will have potentially hundreds of these agents in, you know, some sort of orchestration framework, and, you know, we could see cascading failures occur. We're gonna have to see a whole development of control technologies to, like, cut off, maybe kill, if you like, rogue agents.

Sounds like a William Gibson.

Well, concept that comes to mind with all of this and at some point when I realized the potential of small agents working collaboratively together, science fiction has had the concept of nanobots for quite a long time and I think that what pops into my head is that instead of physical nanobots, these agents are like digital nanobots. The potential for for both goodness and badness is mind blowing completely because of yeah. Enough said.

Yeah. The multiplicative

to that old TV show Chuck two point o. Sorry.

Yeah, and it's hard to we can't ignore the fact that, you know, as cybersecurity professionals, for all these years that we might have been doing this, we've been dealing with deterministic technology, and now the world has kind of flipped on its head, and we're suddenly in this paradigm shift and dealing with nondeterministic technology. And autonomy and nondeterminism flies in the face of good security, so we're gonna have to find this new balance, right, in order to be able to manage our world. And I think I I do think information security practitioners and professionals are going going to have to regear, of course, and we're all in the process of doing that, and we will find that we have a big role to play in consulting on risk mitigation related, especially to agentic multi agent systems.

Yeah, indirect prompt injection is gonna be the number one type vector of 2026.

That's another way to that's another sub prediction, if you like, that goes right along with all the all that statement. Alright. Any other closing thoughts before we close out the episode leading us into 2026? Alright. I'm gonna take as yeah.

I'm just happy that we made it through the whole episode to talk about, you know, AI and jobs.

Right. Let's not talk about AI. Beat you to death. Actually, I'm an optimist on that topic. I'll just say right now, I think AI is going to be an enormous multiplier to demand for skilled individuals in that area. So it doesn't mean you're gonna lose your job. It means that you need to reconsider what you're doing and how you're doing it. That's what the technology puts in front of us.

Did the personal computer take everyone's job? No. Mean, I there were predictions about that at the time. I remember I was alive, right, at the time, which makes me old. And then the same thing for the Internet.

That's right. Alright. We should close it out. So thank you everybody for joining us for this episode. I wanna wish everybody happy holidays. Be safe out there. Enjoy some family time, and we will certainly be seeing you again in 2026 for more episodes of AI Security Ops. Keep on prompting. Be safe out there. We'll see you next time.