Jason Sabin, CTO of DigiCert Inc. Passionate about digital trust, including digital certificate management for web, device & user identity.

getty

Agentic AI has arrived. Enterprises are deploying autonomous agents to retrieve information, interact with systems and execute tasks with minimal human intervention, creating an urgent need for identity, authorization and trust.

Adoption is accelerating as the proverbial wheels are falling off the bus. An estimated 78% of IT and security decision-makers have experienced AI-related incidents or identified AI-related vulnerabilities. Yet while organizations rush to deploy agentic AI, they don’t completely trust it. Many still rely on a human-in-the-loop approach.

That reliance will fade over time and, I believe, disappear entirely. Agentic AI’s pace and shift away from human oversight is driving the demand for proven infrastructure like public key infrastructure (PKI).

Why PKI Works And Will Continue Working

Deploying AI agents requires authentication so you can know everything they are doing. But to know what AI agents are doing, they need identity.

PKI lets organizations issue digital certificates that authenticate users, devices or services. It can give agentic AI that same foundation of identity and trust.

Unlike emerging approaches, PKI has proven itself at global scale over decades, underpinning trust across the internet and machine identity in IoT. As organizations deploy AI agents faster than IoT ever grew, extending that same framework to agentic AI is a logical step.

Why OAuth Falls Flat Without Human-In-The-Loop

The past few months have seen a spike in OAuth identity federation for AI agent authentication.

That works for now, while human-in-the-loop still governs agentic AI. But once humans are out of the loop, OAuth stops working: There’s no one left to enter a username and password to grant access. OAuth isn’t even a great option for human-in-the-loop, either. Before long, humans will tire of reauthenticating AI agents. We need solutions now, and PKI is ready today.

PKI And Private MTCs: Two Paths That Work

Before embracing PKI, the IoT industry tried blockchain, and things fell apart pretty quickly.

Engineers had to rebuild protocols and standards for interoperability, which added complexity and slowed progress. PKI is open source and doesn’t have the same issues. To enable SSL or TLS-encrypted communication, for example, existing PKI libraries are freely available and already built into most devices and operating systems. PKI is not the only approach worth building on.

Private Merkle Tree Certificates (MTCs) offer a complementary option, particularly for organizations running closed, high-volume identity systems. MTCs use hash-based data structures to batch-issue certificates at scale, reducing the overhead of traditional certificate authority chains. In private deployments, they work alongside PKI.

Both approaches share the same foundation: giving AI agents verifiable identity without a human authorizing every transaction. PKI remains the more mature and broadly supported choice. Private MTCs are worth consideration as agentic deployments grow in scale.

MCP For The Win, But Not With OAuth

In 2024, Anthropic introduced Model Context Protocol (MCP), which sits in front of APIs, databases and file systems to allow AI agents to gather information from those sources.

MCP has been in use for roughly a year and has become the default way AI agents gather data. But there’s also a widespread recognition that it lacks strong identity and security.

OAuth 2.1 is now the mandated authentication framework across MCP server implementations.

For now, it works. A human authorizes a token, the agent gets access and the cycle repeats on whatever schedule the organization sets.

The problem is scale. OAuth was built around human authorization, and every token still needs a person to grant it. That dependency breaks down at the speed and volume agentic AI is heading toward.

The Formula For Trusted AI: AIC + PKI, Whatever The Access Point

Just as HTTP shifted to HTTPS, MCP is one access point that needs to mature, but it isn’t the only approach.

While it took the internet pioneers about three years to make that shift, AI is not waiting. Agents are already moving past structured protocols entirely. They access APIs directly, bypassing MCP. They run inside browser extensions, logging in to websites and executing tasks behind users’ screens, acting on behalf of individuals and organizations across surfaces never built for agent access.

MCP is the more structured path and where much of today’s security conversation is focused. But the identity program is the same no matter how an agent gains access. An API call needs verifiable identity, too.

That is why PKI and the AI Identity Certificate (AIC) are not an MCP fix. They are the identity layer that follows the agent wherever it goes.

Here’s how it works: Say you want to grant an AI agent access to a CRM system, a direct API or a resource behind a browser extension. Where MCP is in use, certificate-based message signing is already built into the protocol and the AIC can serve as that certificate. The same AIC works for mTLS, so one credential covers both the protocol layer and mutual authentication.

No matter how the agent connects, through MCP, a direct API call or a browser extension, the authentication mechanism is the same: mTLS. The agent connects to the target resource, which responds with its own certificate, and both sides validate each other.

This confirms the agent is who it claims to be and that the resource it’s reaching is legitimate, regardless of how the connection is made.

Digital certificates also allow for immediate revocation: You can check whether an agent’s identity, or the resource it connects to, is still valid and cut access if it isn't. You can log requests an agent sends across every access point, then replay those decisions to see how, when and why they happened.

The Bottom Line

AI is the top investment priority for 71% of global CEOs, according to KPMG. Spending on agentic AI could reach $155 billion by 2030, per Bank of America.

Secure and autonomous agentic AI is on the horizon as the sun is setting on human-in-the-loop. Businesses need the right solutions that let them act now, and PKI is a tried-and-true technology that empowers them to do both.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?