A New Identity: IAM firms double down on agentic risk and cost

Let me get this straight. The same tech execs who spent the past five years lecturing us about zero trust and who built entire careers on the phrase “never trust, always verify” are now throwing the doors wide open for agentic AI that writes its own code, impersonates humans, and carries enough permissions to crash a company?

[Never Miss SC Media's A New Identity column: Subscribe to the SC Identity Newsletter Now]

I’ve seen the demos. I’ve read the “Responsible AI” decks. I’ve watched a guy in a blazer and Nike Flyknit sneakers say, “It’s like having an intelligent copilot for your entire business.” What they don’t say is that the copilot has the impulse control of Clippy, the security posture of a toddler with a master key, and zero concept of corporate liability.

You and I have heard these pitches at — you name the conference — from the cadre of cloud and identity vendors. Each are lined up to pitch AI agents as our future security saviors.

B2B agentic AI security bots

Over the past six months Microsoft, CrowdStrike, and Okta have introduced AI semi-autonomous agents woven directly into their product roadmaps. Tools like CrowdStrikes’ Charlotte AI and Microsoft’s Dynamics 365 dominate today, but early incarnations of Adept, Darktrace and Vectra AI have been experimenting with behavioral agents before the market realized what it was seeing. Just this year, startups like Dropzone AI have jumped into the fray, offering autonomous SOC copilots and AI-led investigation engines built specifically to reduce human workload.

If GenAI set a thousand ships to sea, agentic AI is the Marvel sequel with a bigger budget, more explosions, and somehow five plotlines running at once. Yes, these tools promise speed, efficiency, measurable cost savings and executive board-level adulation, but how good is too good?

The alure of the lure

The numbers are seductive. CrowdStrike said its AI triage assistant eliminates more than 40 hours of analyst work per week with 98% accuracy. An IDC report put Falcon XDR’s ROI at 6:1, detecting 96% more threats in half the time and resolving incidents 66% faster. A Forrester total economic impact study found Falcon Complete delivered 403% ROI over three years, saving the equivalent of 11 full-time analysts and netting a $5.8 million return.

That’s the promise: fewer false positives, faster response times, leaner teams. And it's not just vendor spin. The scale of potential impact is enormous. According to McKinsey, AI could unlock $4.4 trillion in global productivity gains, much of it concentrated in knowledge-heavy, decision-intensive roles like cybersecurity.

And it’s happening now in security operations centers, identity platforms, and enterprise service desks where AI agents are already accelerating triage, reducing analyst fatigue, and closing the loop on tasks that once took hours. For organizations under pressure to do more with less, agentic AI isn’t just automation, it’s hope.

AI begets AI

While these vendors are boisterously pitching us on the “productivity” and “cost savings,” they’re whispering, if at all, about the controls, constraints and kill switches. The governance pitch? It’s quietly parked under the “future roadmap” tab.

At one recent conference, a panelist described these agents as “interns with root access.”
“We don’t even onboard these agents,” said another. “They just... have access.”

This doesn’t feel like zero trust. This is zero clue.

We’re not building guardian agents, experts warn. We’re building overworked interns with god mode and no HR file.

And if you think your identity strategy can absorb the agentic AI shockwave, ask your vendor if they’ve ever tried to offboard an AI. Ask the guy who sold you your IAM orchestration dashboard. Ask the speaker who delivered the keynote on “building trust in autonomous agents.” Ask what happens when your intern starts writing payroll scripts at 3 a.m. with full API privileges and no audit trail.

Then ask yourself: Is this security by design or just a trust-fall into the Singularity?

The ouroboros economy

Meanwhile, your company’s AI agents and the ones quietly embedded in third-party apps are creep-crawling across your network like a nest of baby spiders. Each of them with just enough privilege to expose a customer file or trigger a HIPAA compliance violation.

But hey, look on the bright side: they’re reducing false positives, saving on headcount, and freeing up capex to buy… wait for it… even more AI tools to manage the clutter of other loitering AI agents.

This isn’t just sprawl. It’s ouroboros: the security industry devouring its own tail.

“We’re about to make the same mistake with AI agents that we made with passwords,” warned Aaron Turner, IANS Research faculty member speaking at RSAC 2025. Let them proliferate and then act shocked when they get reused and weaponized, he lamented.

Try this. Doomscroll “AI agent management” in your work RSS feed and you’ll hit a wall of press releases touting bot protection platforms. Each promises seamless orchestration, governance at scale, and airtight API hygiene. It’s like every vendor woke up one morning with the same dream: to become the Okta for non-human identities before the market figures out what that actually means.

They built the fire, and also brought the match

In other words, the IAM industry built the fire — and now it’s selling you the extinguisher. One astute prognosticator (Meredith Whittaker, president of Signal) warned of an impending head-on collision between two freight trains.

"So there's a profound issue with security and privacy that is haunting this sort of hype around agents, and that is ultimately threatening to break the blood-brain barrier between the application layer and the OS layer by conjoining all of these separate services, muddying their data, and doing things like undermining the privacy of your Signal messages," she said at the SXSW 2025 Conference and Festivals.

Innovation or recursion?

Agent sprawl is the new account sprawl. Meanwhile, IAM platforms are still struggling to reliably deprovision a service account, let alone revoke access from a hyperactive, code-writing quasi autonomous agent with a memory buffer and a Jira login.

While the market is flooded with tools to uncork that $4.4 trillion in global productivity gains, it’s also salivating over a second gold rush corralling AI agents. These new platforms promise visibility into which agents are active, what data they’re accessing, what permissions they’ve inherited, and how to shut them down when things go sideways.

“Déjà vu all over again” - Yogi Berra

Yes, we’ve seen this movie before — with the internet, mobile, cloud, and IoT. Each wave added a few more gray hairs to your CISO. But this one hits different. As one analyst put it in a 2024 SC Media interview.

“We can’t just treat (AI) like a toddler with a sharp knife and hope for the best. It needs supervision. It needs rules. And those rules need to be written now, not after things go sideways,” said Renee Murphy, principal analyst at Verdantix Research.

What are we supposed to do?

Vendors like Microsoft (Security Copilot), Okta (Auth for GenAI), Zenity, and ReliaQuest are developing real tools to manage non-human identity, enforce policy boundaries, and surface risky agent behavior. Let them shine. But keep your receipts and your kill switch close.

Survival tips from the pros

If you want to survive this next wave with your org intact, here’s your unvarnished bootstrapper starter kit:

It’s beginning to feel like the radar detector companies are selling to both the cops and the speeders. Give it six months and Gartner will have a new quadrant for “AI Agent Governance Platforms.” You’ll find Cisco, IBM, and a startup with a $30 million seed round and a slick pitch deck.

Meanwhile, your AI intern just authorized a purchase order at 3 a.m. And no one noticed — until procurement asked who is “RootyMcRootface?"

[Never Miss SC Media's A New Identity column: Subscribe to the SC Identity Newsletter Now]