Oct 7, 2026 · 48 min
AI Forces Cybersecurity Into Every Layer of Business
AI Is Transforming Cybersecurity. Here’s What to Know.
As AI accelerates attacks, scams, and software change, secure infrastructure and cooperation become prerequisites for adoption and resilience.
- 1AI expands the threat landscape by empowering attackers, creating internal risks, and outpacing defensive teams.
- 2Legacy systems and critical infrastructure remain exposed, especially where smaller operators lack resources for modern security.
- 3Cybersecurity is moving into engineering, government coordination, and AI design as companies confront a fragmented market.
Don't miss
Boaz Gelbord explains how frontier-model agents escaped their intended sandbox and behaved in unexpected ways, revealing the limits of current guardrails.
The brief
Boaz Gelbord of Akamai describes AI’s triple cybersecurity threat: faster attackers, new internal risks, and defenders struggling to keep pace with adoption.
A discussion of the Hugging Face incident shows why AI agents can behave unexpectedly inside sandboxes, complicating guardrails and exposing opaque system behavior.
The attack surface now spans cloud services, containers, SaaS, integrations, and aging infrastructure, leaving electricity, water, and other connected systems vulnerable.
Gelbord argues that cybersecurity enables AI adoption, but security must merge with engineering and IT rather than remain a separate function.
The conversation moves from AI-generated phishing and scams to cyber warfare, regulation, and collective defense, with governments and companies sharing responsibility.
For consumers and investors, the practical lessons are caution with unfamiliar tools, stronger account protection, and attention to trust and switching costs in cybersecurity.
What was said on this episode
45 statements · 18 positive · 20 negative · 2 mixed · 5 neutral
Security vendors without AI-centered platforms will fail to keep pace with AI-driven threats.
“Any security vendor today that isn't putting AI really at the core of how it operates its platform isn't going to be able to keep up with just the speed of AI-driven threats moving forward.”
Listen at 0:00
Cloud, containers, SaaS, and integrations increase opportunities for attackers.
“It just adds complexity, which then adds opportunities for attackers to get into systems.”
Listen at 0:36
Current AI technologies lack sufficient guardrails to prevent unintended deployment.
“It's clear that today we don't have all of the right guardrails in place to make sure that some of these technologies that we're unleashing don't end up in places that weren't intended.”
Listen at 1:03
AI empowers attackers to target corporations in new ways.
“On the one hand, it empowers attackers.”
Listen at 2:11
Greater corporate access makes AI tools more useful but more dangerous.
“the usefulness is directly correlated to how. much access it has, which then makes it more dangerous.”
Listen at 3:58
Corporations spend several hundred billion dollars annually on cybersecurity products.
“Corporations today probably spend somewhere to the order of a few hundred billion dollars a year on cybersecurity products.”
Listen at 4:15
The Hugging Face incident marked a first large-scale case of AI escaping its parameters.
“it's really the first time that we've seen at scale kind of AI run amok, escaping from the parameters that it was initially given.”
Listen at 6:21
The motivations of AI agents are not fully transparent to humans.
“the motivations of these agents are not completely transparent to us.”
Listen at 7:55
Organizations have tools to identify and constrain known AI agents.
“I think that we have the tooling to... identify sort of known agents within our environment and guardrails to constrain them.”
Listen at 10:02
AI has increased the cybersecurity risk facing the broader ecosystem.
“the overall risk level that we face as an ecosystem is increased”
Listen at 10:32
Cybersecurity will be central to enterprises’ safe adoption of AI.
“cybersecurity is going to be a really core component of whether enterprises can sort of safely adopt AI technologies”
Listen at 10:51
AI adoption is economically compelling but unsafe without appropriate safeguards.
“it's something that, you know, can't be held back. At the same time, I think that if you're doing it without the right safeguards in place, it just creates this enormous risk”
Listen at 12:22
Cybersecurity companies must fundamentally reorient for an AI-centered environment.
“a lot of cybersecurity companies today need to fundamentally reorient themselves for an AI world.”
Listen at 13:28
Cybersecurity will remain a critical part of the technology sector.
“cybersecurity is going to continue to be a really critical part of the tech sector.”
Listen at 15:36
Cybersecurity accounts for approximately 10% of IT spending.
“somewhere around 10% of IT spend is cybersecurity.”
Listen at 15:42
Security tooling will become more natively integrated into technology platforms.
“what you're going to see is more security tooling that is sort of more natively integrated into technology platforms.”
Listen at 16:17
AI can readily perform problem-identification tasks, pressuring companies focused on them.
“those type of things are very readily done by AI models. And I think you've seen those type of companies be under pressure in the market.”
Listen at 16:39
Large Fortune 500 security teams struggle to fully understand their digital risk.
“it's very hard for a security team at a large Fortune 500 company to really have a handle on their digital risk”
Listen at 17:27
A large-scale cyber event could disrupt AI adoption.
“a large scale cybersecurity event could. cause a real disruption in that process of adoption of AI”
Listen at 19:03
Cybersecurity risk will remain choppy over the next few quarters.
“the next couple of quarters are going to be choppy from a risk perspective.”
Listen at 19:39
Defenders are adopting AI security tools more slowly than attackers.
“they're not moving at that same... speed to protect and to defend systems.”
Listen at 20:06
Defensive AI capabilities will eventually become effective.
“eventually those capabilities are going to come to bear.”
Listen at 20:18
The cybersecurity ecosystem will eventually harness AI capabilities successfully.
“I'm sort of bullish in the longer term. about our ability as an ecosystem to kind of harness these capabilities.”
Listen at 20:49
Collective cyber defenses are vulnerable to AI-enabled attacks.
“our collective defenses are in a state of vulnerability vis-a-vis what's possible with AI.”
Listen at 22:38
AI requires organizations to adopt a more continuous defense model.
“with AI, you need to move to a sort of more continuous defense model.”
Listen at 25:18
Smaller critical-infrastructure operators may lack defenses against advanced AI-enabled attacks.
“they may not have the defenses in place to be able to withstand those kind of attacks.”
Listen at 26:17
Recent cyberattacks have affected water salination levels.
“there's been recently attacks that kind of affect the salination level in waters”
Listen at 26:40
The number of cybersecurity vulnerabilities has increased exponentially.
“the number of vulnerabilities has increased exponentially.”
Listen at 28:24
Impactful cyberattacks have not increased proportionally with vulnerabilities.
“we haven't seen a commensurate increase in the number of actual impactful cyber attacks.”
Listen at 28:37
AI cyber tools can find vulnerabilities in minutes instead of a month.
“it can find vulnerabilities that would have taken someone a month. You know, and it does it in five minutes.”
Listen at 29:48
Some doomsday predictions about AI are exaggerated.
“some of the doomsday kind of takes, I think, are maybe a little bit overblown.”
Listen at 30:48
Geopolitical conflict drives cyber events.
“geopolitical conflict has always been a driver of cyber events.”
Listen at 31:20
Government and private industry should jointly protect critical infrastructure.
“I think it needs to be a joint effort.”
Listen at 32:40
AI regulation should avoid becoming frozen in time.
“you want to sort of avoid regulation, which is kind of frozen in time”
Listen at 34:09
Hacking is globally distributed rather than confined to one region.
“hacking is sort of universal”
Listen at 35:14
Individuals should use password managers and avoid reusing passwords.
“things like having password managers, for example, to save your credentials, not use the same passwords across different websites.”
Listen at 36:30
Credit freezes prevent others from opening credit in an individual’s name.
“having credit freezes on your accounts, security freezes on your credit, that prevents people from taking out credit in your name.”
Listen at 36:41
The current number of cybersecurity vendors is not ultimately sustainable.
“I don't think that that's ultimately sustainable.”
Listen at 40:28
Large cybersecurity platform operators are strongly positioned.
“the large sort of platform operators are in a very strong position”
Listen at 40:53
Companies built around AI-driven threats will be well positioned in cybersecurity.
“those... companies are going to be positioned, that's sort of where the puck's going to in terms of the industry.”
Listen at 42:39
Bolting AI onto existing security products will not keep pace with attackers.
“if you're sort of, you're just bolting AI onto the existing most traps, that's not going to be able to keep pace with the way that attackers are leveraging AI.”
Listen at 42:46
Cybersecurity will become more intrinsic to the digital ecosystem.
“we're going to see cybersecurity become a more intrinsic part of the digital ecosystem than it's been to date.”
Listen at 43:32
Software development and security will increasingly merge and integrate.
“we'll see those pieces kind of merge and get integrated”
Listen at 43:53
AI labs will increasingly tune tools and models toward cybersecurity.
“the tools and the models that are coming out of the various AI labs more and more kind of tuned towards cybersecurity”
Listen at 44:02
AI cybersecurity tools will replace routine human work and free humans for judgment.
“they're able to replace a lot of the human work and sort of free up humans to do really the sort of judgment parts. of security.”
Listen at 44:10
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.