Agentic AI Is Doing What a Thousand Breach Reports Couldn't: Getting Boards to Open the Checkbook

Key points

• Consulting firm Bain & Co. Has warned that organizations may need to double or triple security investment to keep pace with advanced AI systems.
• Agentic systems operate across dozens of enterprise interfaces at machine speed, with credential and session-token exposure that existing PAM tooling was never designed to handle.
• AI-generated code is introducing authentication flaws and insecure dependencies faster than security review cycles can absorb them.
• CISOs say the budget pitch that survives board scrutiny frames security as the governance layer that lets the business scale AI safely, not as insurance against catastrophe.

For roughly twenty years, CISOs have rehearsed the same budget pitch. A breach happens, urgency spikes, a few dollars shake loose, and the moment passes. AI may actually break the cycle.

The arrival of frontier models capable of autonomous offensive operations, combined with the spread of agentic AI inside enterprises, has produced something rare: board-level urgency that doesn't require a headline-making incident to sustain itself. Bain & Co. Has warned that organizations may need to double or triple security investment just to keep pace with the operational complexity these systems introduce.

That math is landing differently than past threat briefings. "Because of [recent AI developments], there's been this realization that we haven't enabled AI as much as we need to in security," said Nate Rollings, CISO at Zafran Security, speaking at the CSO Cybersecurity Awards and Conference in Nashville. "We're seeing buy-in from the top down."

Should identity teams be worried?

The identity implications are concrete. Agentic systems need credentials, session tokens, and API keys. They chain calls across dozens of internal interfaces at sub-second intervals with minimal human oversight. Conventional privilege models assume a human sits somewhere in the loop. Agents don't.

"I've got a single potential agent that can go interact with all 50 interfaces available in the company in a sub-second," said Bernard Brantley, CISO at Corelight. Give every employee three agents and the addressable identity surface grows by orders of magnitude overnight. Refresh-token rotation policies and SCIM provisioning workflows simply were not designed for this pace.

Our June reporting on a poisoned npm package showed exactly how exposed agent credential stores already are: malicious artifact builds harvested refresh tokens while the source repo looked clean.

Diana Kelley, CISO at Noma Security, put it plainly: "We're now protecting a decision and automation layer." Agentic identities need the full lifecycle treatment, logging, behavioral baselining, anomaly detection, deprovisioning, that mature organizations apply to privileged human accounts.

AI-assisted coding compounds the exposure. Developers shipping AI-generated code at accelerated velocity are introducing authentication flaws and insecure dependency chains that security review cycles cannot absorb. More surface, less oversight.

Should boards be spending more?

Not everyone is ready to declare a spending revolution. Ian Thornton-Trump, CISO at Inversion6, offered the necessary counterweight: vague AI fear is not a business case. "Waving the flag of AI is the wrong answer," he said. The pitch that survives scrutiny frames security as the governance layer that lets the business scale AI without losing visibility or control.

Brantley put it precisely: "The increase in cyber budget should be oriented toward delivering business value with respect to the current or strategic AI goal." Security as operational prerequisite rather than insurance policy is the version of the argument that actually closes.