This is the second episode of Agent Eye-Openers, a series of conversations that delves into the decisions, discoveries, and wake-up calls defining the AI era.
You can't hold an AI agent accountable. If it accidentally exposes sensitive customer data, you can't fire it—it is, at the end of the day, still just a machine.
For Micah Spieler, Chief Product Officer at Strike Graph—an AI-native compliance platform—this realization was a wake-up call, signaling that organizations are now dealing with an entirely new identity type that requires a new approach to security.
This was especially true for Strike Graph, which houses sensitive information about how organizations manage regulatory requirements. So when the company started exploring AI agents and large language models, it faced a critical question: How do you unlock the power of these tools without putting customer data at risk?
A blueprint for managing ‘hungry agents’
For Strike Graph, it started with acknowledging that just because AI agents can parse troves of data, it doesn’t mean they should. “AI agents are hungry. They'll take whatever information you give them because that's how they operate and perform the task the most efficiently," Spieler explains. If you grant an agent access to your entire database to complete one task, it will use all of it—not out of malice, but by design.
This created a real tension for Strike Graph's customers. They wanted to harness the benefits of MCP servers and LLMs, but they were understandably nervous about exposing sensitive compliance data to these new tools.
To confidently manage these agents, Strike Graph leans into Okta’s blueprint for a secure agentic enterprise, which requires organizations to answer three questions: Where are my agents? What can they connect to? And what can they do?
By leveraging Auth0 for AI Agents, Strike Graph can confidently answer all three:
Where are my agents? Strike Graph hosts its own models entirely within its own firewall to maintain full control over customer data. Auth0 helps secure this boundary, confirming that agents only run within their trusted ecosystem without relying on unsecured third-party APIs.
What can they connect to? Using Auth0, Strike Graph added a new authentication layer to its Model Context Protocol (MCP) servers, allowing its customers to set fine-grained access controls and helping ensure agents only connect to authorized data sources.
What can they do? Auth0's ability to differentiate between human and AI agent traffic patterns gives Strike Graph robust logging capabilities. Non-technical users, such as chief compliance officers, can easily audit the system to see exactly which agents accessed what data and when.
This level of visibility and control translates into customer confidence. When compliance-sensitive customers ask tough questions about data management, Strike Graph can provide the answers, helping customers say "yes" to AI features faster.
Speed without shortcuts
For Strike Graph, security is paramount but it also can’t slow product teams down. “It's critical that our engineering team doesn't have to spend extra cycles building the authentication layer," Spieler explains. By utilizing Auth0 to provide the authentication layer for their AI agents and MCP servers, Spieler can free up his team’s time to focus on piloting new features and products for customers.
Offloading this sensitive auth layer also allows Strike Graph to maintain go-to-market velocity, turning internal concepts into production-ready prototypes in a matter of days."We are able to take an idea, put it into production for customers, and get their feedback all within one sprint," Spieler says.
The rising trend of chaining models
The need for a secure foundation becomes even more important as AI development evolves beyond single models into interconnected chains. Rather than relying solely on massive frontier models, Spieler notes a rising trend: "chaining" together smaller, finely tuned models to handle complex tasks in sequence. By passing data from one specialized step to the next, these smaller models run with a lower footprint and higher accuracy on properly completing specific tasks.
However, chaining multiple models means data is passed around much more frequently. In an ecosystem where MCP servers are communicating with each other and hundreds of different models, agent authentication is paramount. Organizations have to think critically about how they use authentication to ensure data remains encrypted and is only accessed by the right AI agents, at the right time, and for the right purpose.
Managing this complexity requires a dedicated identity layer to mediate between these chained models. As Spieler emphasizes, “You have to secure the hand-off. If you don't verify the identity of the agent at every step of the chain, you’re leaving yourself open to risk.” By authenticating each interaction, with solutions like Auth0 for AI Agents, organizations can help ensure sensitive context is passed safely downstream, preventing un-permissioned agents from going rogue or accessing unauthorized resources in the workflow.
Unlocking the ‘wow’ moments
Solving identity for AI agents is critical for preventing a catastrophic security breach, but it's about unlocking the technology's full potential to delight users.
As Spieler points out, when customers see that their data is treated with absolute privacy and secure controls, they’re more comfortable with experimenting and trying new features. "To watch the wow on their face when our AI security assistant connects the dots on a really complex compliance question is pretty incredible," Spieler says.
By establishing a trusted authentication layer from day one, companies can focus on creating those powerful AI experiences that delight customers and drive product innovation forward.
Watch the full video above to hear more of Micah's insights on AI agents, compliance, and emerging authentication challenges.
