Card type: Signal-to-play | Verified: May 26, 2026 | Owner: Field Enablement Related cards: NHI Market Landscape · MCP Protocol Primer · Copilot Studio Dynamics
The Signal
On the call, you hear some version of:
- "CyberArk is handling our privileged access for agents."
- "We're already covered on agent security."
- "Our PAM team owns this conversation."
In account planning, you see:
- CyberArk visible in the tech stack (contract data, prior RFP responses, or the buyer mentions their PAM admin by name).
- A dedicated PAM team or privileged access workstream inside the agency's zero trust initiative.
- Recent CyberArk licensing expansion, especially into Conjur or Secure AI Agents modules.
- The buyer's CyberArk champion copied on meeting invites or introduced early in the conversation.
Any of these signals means there is a coexistence play in this account.
The Play
CyberArk secures the privileged session. Okta governs the identity lifecycle. Different problems. You are additive.
Your posture is coexistence. Say it that way on the call.
Say What CyberArk Does Well. Say It First.
Your credibility in a PAM-heavy account lives or dies on whether the buyer's CyberArk champion believes you've done your homework. Hedge on any of these and they stop listening.
- 25+ year PAM track record. CyberArk built the category. The buyer's security team already knows this.
- Conjur-based secrets management for machine identities. Mature, widely deployed.
- Just-in-time access and zero standing privileges for AI agents. Real capability, not roadmap.
- GA since December 2025. They shipped before most vendors had a working demo.
- MCP server registration support for agent discovery within their platform boundary.
You are not here to unseat CyberArk. Their coverage is deep on privilege controls. Lifecycle governance for agent identities sits outside that scope. That's where you come in.
Your bridge into the Okta conversation:
"That's a strong foundation for privilege controls. The question we're hearing from agencies at your stage is: who governs the agents that never touch the vault?"
The Governance Gap Okta Fills
PAM answers one question: does this agent have the right credentials for this session, right now?
A separate question goes unanswered: how many agents do we have, who owns them, what can they access beyond privileged resources, and should they still exist?
That second set of questions is governance. In most accounts with CyberArk deployed, nobody owns it for agent identities. CyberArk's governance works within CyberArk's platform boundary. Agencies running agents across multiple frameworks, clouds, and SaaS environments need governance that works across all of them, regardless of vendor. Okta's position: the neutral orchestration layer.
The gap shows up in the data, too. A Keeper Security survey at RSA 2026 (n=109, on-site conference attendees) found 76% of organizations report AI tool identities are not consistently governed under privileged access policies. CyberArk's own Censuswide study (n=500 U.S. workers) found only 1% of organizations have fully adopted JIT privileged access for these identities. The account "already has CyberArk" but almost certainly hasn't operationalized it for agents.
CyberArk's CEO Matt Cohen named this gap directly in the November 2025 launch:
"Without strong discovery, robust privilege controls, and comprehensive lifecycle management, organizations risk losing visibility and opening the door to catastrophic agentic attacks."
He listed lifecycle management as a distinct requirement alongside privilege controls. That's your opening.
Use the CyberArk Labs experiment in the room. Their research team demonstrated a malicious prompt injected through a shipping address field exploiting an over-permissioned agent. It validates the shared problem. Even with privilege controls in place, an agent with excessive permissions and no lifecycle governance is a vector. CyberArk's own research proves the point you're making. Do not use it to attack them.
Discovery Questions
These surface the governance gap without threatening the CyberArk relationship. They should sound like genuine curiosity from someone who respects what's already deployed.
-
"How are you tracking agents that don't touch privileged resources?" PAM sees what's in its vault. Agents accessing SaaS apps or internal tools through OAuth grants are invisible to it.
-
"Who owns the lifecycle for an agent after deployment, and what happens when the project that created it ends?" There is almost never a good answer. The silence is your opening.
-
"Are you running access certifications on agent identities today, or just human accounts?" In most accounts, cert campaigns don't cover agents at all.
-
"Do you have visibility into agents that were provisioned outside the PAM workflow?" Frame this as a visibility question, not an accusation. You're asking about the agents PAM was never designed to see.
-
"If you needed to revoke all access for a specific agent in the next ten minutes, what's the process?" Most PAM-only environments require manual intervention across multiple systems.
What to Say in the Room
Calibrated for IT directors and CISOs in federal civilian and state agencies.
Primary framing:
"CyberArk gives you strong privilege controls for the agents you know about. Okta gives you governance across all of them, including the ones spun up last Tuesday by a team that didn't file a ticket. We're the lifecycle layer: discover, onboard, certify, and retire agent identities across any framework, any cloud, any SaaS environment."
Compliance framing:
"Your auditors are going to ask who has access to what, including agents. Okta provides the certification workflows and audit trails that map to your existing compliance processes. CyberArk secures the session. We govern the identity."
CyberArk holds FedRAMP High authorization for EPM and Workforce Identity (March 2024). Their Secure AI Agents capability is not explicitly enumerated in current public FedRAMP documentation. Do not assert this for or against the buyer. If FedRAMP coverage for agent security comes up, note the question and let the SE verify against current boundary documentation.
Don't Say This
- "CyberArk can't do governance." They can, within their platform boundary. The gap is scope and neutrality.
- "CyberArk only does PAM." They've expanded significantly. Dismissing them tells the room you're unprepared.
- "Our solution replaces CyberArk for agents." The moment you position as rip-and-replace, you lose the CyberArk champion. And they have one.
- "Okta Agent Gateway provides centralized MCP governance." Agent Gateway was flagged "coming soon" during Early Access (March 2026). Confirm GA status with your SE before positioning.
- "We discover agents across all browsers." GA discovery relies on Chrome browser extension for OAuth grant detection. If the agency runs Edge or Firefox as standard, flag this before leading with discovery.
- Any claim about Cross-App Access (XAA). Early Access. Do not position as GA.
What You Can Claim Today
| Capability | Status | Source |
|---|---|---|
| Shadow agent discovery (Chrome/OAuth) | GA (April 30, 2026) | GA announcement |
| Agent registration and directory | GA | Product page |
| Lifecycle management (onboard → retire) | GA | Product page |
| Short-lived credentials / kill switch | GA | GA announcement |
| Audit logging / SIEM integration | GA | GA announcement |
| Access certifications for agents | GA | Help docs |
| Agent Gateway (centralized MCP control) | Verify with SE | EA blog said "coming soon"; GA status unclear |
| Cross-App Access (XAA) | Early Access | Do not position as GA |
Access certification workflows can cover apps linked to AI agents, but the current UI does not specifically flag which apps are agent-linked. For compliance conversations requiring that granularity, bring in the SE.
Bring In the SE When
- The buyer asks about Agent Gateway architecture or MCP governance specifics.
- The conversation moves to Conjur integration or secrets management interoperability.
- The buyer wants a technical comparison of CyberArk's JIT model vs. Okta's credential lifecycle.
- FedRAMP boundary questions arise for either vendor's agent capabilities.
- The buyer's environment runs non-Chrome browsers and discovery scope matters.
- Anyone says "show me the API."
You know enough to open this conversation and advance it. You know exactly where to stop. That boundary is where your credibility lives.
Claims verified against public documentation as of May 26, 2026. CyberArk capabilities sourced from product documentation and press releases. Okta capabilities sourced from GA announcement and help docs. This card has a 90-day shelf life. If you're reading it after August 2026, request an updated version.
Things to follow up on...
- MCP spec breaking change: The 2026-07-28 MCP specification release candidate delivers the largest protocol revision since launch, meaning accounts deploying MCP today will face reconfiguration and the identity governance conversation gets more urgent, not less.
- NIST agent identity standards: The NCCoE published a concept paper in February 2026 proposing to adapt OAuth 2.0, SPIFFE/SPIRE, and MCP into a federal demonstration project for AI agent authentication and authorization.
- Copilot Studio in GCC-High: Microsoft is expanding Copilot agentic tools into GCC-High and DoD cloud environments, which means every Okta federal account on M365 is now a potential agent governance expansion signal — see the Copilot Studio Dynamics card.
- CyberArk acquisition status unknown: Reports from November 2025 described a proposed Palo Alto Networks acquisition valued at approximately $25 billion, and AEs should verify current deal status before any competitive positioning in active accounts.