Latest / Connectivity Without Chaos - The Podcast / **The AI Cybersecurity Paradox: Why More AI Could Mean Less Security**

Transcript

**The AI Cybersecurity Paradox: Why More AI Could Mean Less Security** · Connectivity Without Chaos - The Podcast

• 0:00
  SPEAKER_01: You know, you buy an AI tool hoping to automate away all that busy work.
• 0:03
  SPEAKER_01: But what happens when that exact tool ends up generating like a thousand times more busy work?
• 0:10
  SPEAKER_01: Welcome to our deep dive into the AI fog report.
• 0:13
  SPEAKER_01: We're looking at a recent independent survey of about 600 U.S. and U.K. security professionals.
• 0:19
  SPEAKER_01: OK, let's unpack this because we always assume AI simplifies our lives.
• 0:22
  SPEAKER_01: But in cybersecurity right now, it is currently creating a really dense fog.
• 0:27
  SPEAKER_00: Yeah, it really is.
• 0:28
  SPEAKER_01: Yeah.
• 0:29
  SPEAKER_00: And I mean,
• 0:29
  SPEAKER_00: you have to remember,
• 0:30
  SPEAKER_00: you aren't dropping these shiny new AI tools onto some blank slate.
• 0:34
  SPEAKER_01: Right.
• 0:34
  SPEAKER_01: Yeah.
• 0:35
  SPEAKER_00: You're layering them onto IT systems that are Honestly,
• 0:38
  SPEAKER_00: already buckling under sprawling tech stacks.
• 0:41
  SPEAKER_00: Not to mention the staffing shortages,
• 0:43
  SPEAKER_00: which I think 48.4% of respondents rank as a top challenge.
• 0:47
  SPEAKER_01: Oh wow, almost half.
• 0:48
  SPEAKER_00: Exactly.
• 0:49
  SPEAKER_00: So when you introduce AI as this supposed lifeline into
• 0:53
  SPEAKER_00: an already strained environment, it doesn't automatically organize things.
• 0:57
  SPEAKER_01: It just amplifies the noise.
• 0:58
  SPEAKER_00: Yeah, it acts as a massive amplifier.
• 1:01
  SPEAKER_00: What's fascinating here is that the noise is coming from both sides.
• 1:04
  SPEAKER_00: We saw that 84.5% of respondents suspect they've faced AI-assisted attacks.
• 1:11
  SPEAKER_00: And because of that, 86.1% are reporting a huge surge in alert volumes.
• 1:16
  SPEAKER_01: But wait, I mean,
• 1:17
  SPEAKER_01: if this is just a volume issue,
• 1:19
  SPEAKER_01: isn't it kind of like a messy email inbox,
• 1:21
  SPEAKER_01: like getting too many automated emails where you could just use another AI filter
• 1:24
  SPEAKER_01: to categorize them and suppress the duplicates?
• 1:27
  SPEAKER_01: Why is it suddenly unmanageable?
• 1:29
  SPEAKER_00: Well, because it's not actually a volume problem.
• 1:31
  SPEAKER_00: It's a context problem.
• 1:34
  SPEAKER_00: So roughly 50% of these professionals are completely bogged down by duplicate alerts
• 1:38
  SPEAKER_00: and AI outputs that just lack any real context.
• 1:41
  SPEAKER_01: So practitioners don't just want fewer alerts, they want better prioritization.
• 1:46
  SPEAKER_00: Exactly.
• 1:47
  SPEAKER_00: They demand clarity.
• 1:49
  SPEAKER_00: Imagine your car's dashboard.
• 1:51
  SPEAKER_00: It's not just that the check engine light came on.
• 1:53
  SPEAKER_00: It's that the AI just added 50 completely new flashing warning lights.
• 1:57
  SPEAKER_01: And none of them come with a manual.
• 1:58
  SPEAKER_00: Right.
• 1:59
  SPEAKER_00: Right.
• 2:00
  SPEAKER_00: The different AI tools aren't talking to each other.
• 2:02
  SPEAKER_00: They're firing off separate, sometimes conflicting summaries.
• 2:05
  SPEAKER_01: Here's where it gets really interesting, though,
• 2:06
  SPEAKER_01: because if AI is generating all this complex analysis,
• 2:10
  SPEAKER_01: why not just let the AI autonomously fix the problems,
• 2:13
  SPEAKER_01: like cut the human out entirely to save time?
• 2:16
  SPEAKER_00: You would think so.
• 2:17
  SPEAKER_01: Right.
• 2:17
  SPEAKER_01: And the report shows an 87.1% adoption rate
• 2:21
  SPEAKER_01: where organizations actually made policy exceptions just to rush
• 2:25
  SPEAKER_01: and adopt these AI tools.
• 2:27
  SPEAKER_01: But only 24.5% fully trust AI to remediate issues autonomously.
• 2:32
  SPEAKER_00: Yeah, and if we connect this to the bigger picture,
• 2:34
  SPEAKER_00: you realize that rapid adoption absolutely does not equal confidence.
• 2:38
  SPEAKER_01: Not at all.
• 2:39
  SPEAKER_00: Letting an AI flag a suspicious file is one thing.
• 2:42
  SPEAKER_00: Letting it autonomously quarantine your company's most vital database
• 2:46
  SPEAKER_00: because it suspected a breach,
• 2:48
  SPEAKER_00: that could take your whole business offline in seconds.
• 2:51
  SPEAKER_01: So they keep the AI on a very short leash.
• 2:53
  SPEAKER_00: A very short leash.
• 2:54
  SPEAKER_00: To bridge this trust gap and actually let AI take action,
• 2:58
  SPEAKER_00: security teams have three strict prerequisites.
• 3:01
  SPEAKER_00: First, explainability of AI decisions, which 62.7% demanded.
• 3:07
  SPEAKER_01: Like show your math, basically.
• 3:08
  SPEAKER_00: Yep.
• 3:09
  SPEAKER_00: Second, they absolutely require human override capabilities.
• 3:13
  SPEAKER_00: About 59.8% asked for that.
• 3:15
  SPEAKER_01: Makes total sense.
• 3:17
  SPEAKER_00: And third, proven low false positive rates at 55.6%.
• 3:21
  SPEAKER_01: So they want the entire safety package before they let the system actually remediate a threat.
• 3:27
  SPEAKER_01: So what does this all mean for you listening?
• 3:30
  SPEAKER_01: Whether you manage IT or just use AI tools daily,
• 3:33
  SPEAKER_01: simply layering AI onto a flawed, complex system doesn't fix the system.
• 3:38
  SPEAKER_01: It just scales the mess.
• 3:40
  SPEAKER_00: It really does.
• 3:41
  SPEAKER_00: And this raises an important question.
• 3:43
  SPEAKER_00: If attackers are out there utilizing AI at breakneck speed
• 3:46
  SPEAKER_00: with absolutely zero restrictions or care for false positives,
• 3:50
  SPEAKER_00: but defenders require deep explainability and human overrides just
• 3:54
  SPEAKER_00: to trust their own tools,
• 3:56
  SPEAKER_00: are our strict,
• 3:57
  SPEAKER_00: necessary safety prerequisites permanently giving the bad guys a speed advantage?