⸻ Podcast: Redefining Society and Technology
https://redefiningsocietyandtechnologypodcast.com 

_____________________________

This Episode’s Sponsors

BlackCloak provides concierge cybersecurity protection to corporate executives and high-net-worth individuals to protect against hacking, reputational loss, financial loss, and the impacts of a corporate data breach.

BlackCloak:  https://itspm.ag/itspbcweb

_____________________________

A Musing On Society & Technology Newsletter Written By Marco Ciappelli | Read by TAPE3

August 9, 2025

The Agentic AI Myth in Cybersecurity and the Humanity We Risk When We Stop Deciding for Ourselves
Reflections from Black Hat USA 2025 on the Latest Tech Salvation Narrative


Walking the floors of Black Hat USA 2025 for what must be the 10th or 11th time as accredited media—honestly, I've stopped counting—I found myself witnessing a familiar theater. The same performance we've seen play out repeatedly in cybersecurity: the emergence of a new technological messiah promising to solve all our problems. This year's savior? Agentic AI.

The buzzword echoes through every booth, every presentation, every vendor pitch. Promises of automating 90% of security operations, platforms for autonomous threat detection, agents that can investigate novel alerts without human intervention. The marketing materials speak of artificial intelligence that will finally free us from the burden of thinking, deciding, and taking responsibility.

It's Talos all over again.

In Greek mythology, Hephaestus forged Talos, a bronze giant tasked with patrolling Crete's shores, hurling boulders at invaders without human intervention. Like contemporary AI, Talos was built to serve specific human ends—security, order, and control—and his value was determined by his ability to execute these ends flawlessly. The parallels to today's agentic AI promises are striking: autonomous patrol, threat detection, automated response. Same story, different millennium.

But here's what the ancient Greeks understood that we seem to have forgotten: every artificial creation, no matter how sophisticated, carries within it the seeds of its own limitations and potential dangers.

Industry observers noted over a hundred announcements promoting new agentic AI applications, platforms or services at the conference. That's more than one AI agent announcement per hour. The marketing departments have clearly been busy.

But here's what baffles me: why do we need to lie to sell cybersecurity? You can give away t-shirts, dress up as comic book superheroes with your logo slapped on their chests, distribute branded board games, and pretend to be a sports team all day long—that's just trade show theater, and everyone knows it. But when marketing pushes past the limits of what's even believable, when they make claims so grandiose that their own engineers can't explain them, something deeper is broken.

If marketing departments think CISOs are buying these lies, they have another thing coming. These are people who live with the consequences of failed security implementations, who get fired when breaches happen, who understand the difference between marketing magic and operational reality. They've seen enough "revolutionary" solutions fail to know that if something sounds too good to be true, it probably is.

Yet the charade continues, year after year, vendor after vendor. The real question isn't whether the technology works—it's why an industry built on managing risk has become so comfortable with the risk of overselling its own capabilities. Something troubling emerges when you move beyond the glossy booth presentations and actually talk to the people implementing these systems. Engineers struggle to explain exactly how their AI makes decisions. Security leaders warn that artificial intelligence might become the next insider threat, as organizations grow comfortable trusting systems they don't fully understand, checking their output less and less over time.

When the people building these systems warn us about trusting them too much, shouldn't we listen?

This isn't the first time humanity has grappled with the allure and danger of artificial beings making decisions for us. Mary Shelley's Frankenstein, published in 1818, explored the hubris of creating life—and intelligence—without fully understanding the consequences. The novel raises the same question we face today: what are humans allowed to do with this forbidden power of creation? The question becomes more pressing when we consider what we're actually delegating to these artificial agents. It's no longer just pattern recognition or data processing—we're talking about autonomous decision-making in critical security scenarios. Conference presentations showcased significant improvements in proactive defense measures, but at what cost to human agency and understanding?

Here's where the conversation jumps from cybersecurity to something far more fundamental: what are we here for if not to think, evaluate, and make decisions? From a sociological perspective, we're witnessing the construction of a new social reality where human agency is being systematically redefined. Survey data shared at the conference revealed that most security leaders feel the biggest internal threat is employees unknowingly giving AI agents access to sensitive data. But the real threat might be more subtle: the gradual erosion of human decision-making capacity as a social practice.

When we delegate not just routine tasks but judgment itself to artificial agents, we're not just changing workflows—we're reshaping the fundamental social structures that define human competence and authority. We risk creating a generation of humans who have forgotten how to think critically about complex problems, not because they lack the capacity, but because the social systems around them no longer require or reward such thinking.

E.M. Forster saw this coming in 1909. In "The Machine Stops," he imagined a world where humanity becomes completely dependent on an automated system that manages all aspects of life—communication, food, shelter, entertainment, even ideas. People live in isolation, served by the Machine, never needing to make decisions or solve problems themselves. When someone suggests that humans should occasionally venture outside or think independently, they're dismissed as primitive. The Machine has made human agency unnecessary, and humans have forgotten they ever possessed it. When the Machine finally breaks down, civilization collapses because no one remembers how to function without it.

Don't misunderstand me—I'm not a Luddite. AI can and should help us manage the overwhelming complexity of modern cybersecurity threats. The technology demonstrations I witnessed showed genuine promise: reasoning engines that understand context, action frameworks that enable response within defined boundaries, learning systems that improve based on outcomes. The problem isn't the technology itself but the social construction of meaning around it. What we're witnessing is the creation of a new techno-social myth—a collective narrative that positions agentic AI as the solution to human fallibility. This narrative serves specific social functions: it absolves organizations of the responsibility to invest in human expertise, justifies cost-cutting through automation, and provides a technological fix for what are fundamentally organizational and social problems.

The mythology we're building around agentic AI reflects deeper anxieties about human competence in an increasingly complex world. Rather than addressing the root causes—inadequate training, overwhelming workloads, systemic underinvestment in human capital—we're constructing a technological salvation narrative that promises to make these problems disappear.

Vendors spoke of human-machine collaboration, AI serving as a force multiplier for analysts, handling routine tasks while escalating complex decisions to humans. This is a more honest framing: AI as augmentation, not replacement. But the marketing materials tell a different story, one of autonomous agents operating independently of human oversight.

I've read a few posts on LinkedIn and spoke with a few people myself who know this topic way better than me, but I get that feeling too. There's a troubling pattern emerging: many vendor representatives can't adequately explain their own AI systems' decision-making processes. When pressed on specifics—how exactly does your agent determine threat severity? What happens when it encounters an edge case it wasn't trained for?—answers become vague, filled with marketing speak about proprietary algorithms and advanced machine learning.

This opacity is dangerous. If we're going to trust artificial agents with critical security decisions, we need to understand how they think—or more accurately, how they simulate thinking. Every machine learning system requires human data scientists to frame problems, prepare data, determine appropriate datasets, remove bias, and continuously update the software. The finished product may give the impression of independent learning, but human intelligence guides every step.

The future of cybersecurity will undoubtedly involve more automation, more AI assistance, more artificial agents handling routine tasks. But it should not involve the abdication of human judgment and responsibility. We need agentic AI that operates with transparency, that can explain its reasoning, that acknowledges its lim