Software companies are looking to find new ways to harness AI for an edge. As with any new technology, however, there are caveats to adoption. Identity is among the elements to figure out before AI can be trusted to drive innovation.
One company focused on addressing the identity issue is Okta, which recently discussed its efforts in this episode of the AWS for Software Companies Podcast. Bhawna Singh, chief technology officer for customer identity at Okta, shared the four critical identity patterns needed to build the trust that leads to AI adoption.
The Problem
Everyone is talking about AI, including agentic AI. According to research from leading consulting firm Capgemini, 99% of organizations are implementing AI agents, exploring their potential, or plan to do so within the next six to 12 months. However, numerous challenges must be overcome before these adoption plans can bear fruit—specifically, creating trust within organizations. Research from Salesforce points out that 61% of organizations say they lack trusted data and the security skills that are essential before widespread adoption can occur.
Additionally, there is a catch-22 here: The more AI adoption plans expand, the more risks will likely be exposed, resulting in even more trust concerns. This paradox is particularly important as organizations feel pressured to bring AI capabilities to market faster and with a broader set of features than in a typical initial release.
Agentic AI makes this trust even more challenging and critical, as thousands and even millions of agents interact autonomously with an organization’s data, other applications, and even other agents. You need to have confidence that these agents are doing what they should, with minimal to no human intervention or oversight, before you embark on widespread adoption.
The Solution
Trust fundamentally starts with identity. AI adoption depends on your organization being able to trust that your AI tools and solutions are securely processing data and appropriately acting to complete objectives. Identity is at the core.
This means identity must be part of the design process in your AI implementation. Identity requires four critical capabilities:
- Authentication. Is the agent authenticated, or is it rogue? You must identify the agent in order to authenticate it and install the proper security guardrails.
- API-to-API calls. These calls must be secure at all times. Consider an email agent trying to access your Gmail account. It needs to connect with application data, and both sides need to be able to talk to each other. How do you know that the agent has been authenticated and that you’re dealing with the right agent? To do that, you need tools, such as session tokens and token vaults, that enable secure agent-to-agent communication.
- Asynchronous user confirmation. To prevent rogue decisions, you need asynchronous user approval. Consider an agent that helps you buy and sell stocks; you certainly don’t want the agent to exceed limits that you establish. Instead, the agent should obtain confirmation before acting on your behalf.
- Authorization. Does the agent have the right access to the right data set? For example, Okta has developed its own fine-grained authorization to prevent something like a rogue agent deleting a database.
These capabilities are designed to ensure you have the right controls in place and can establish full traceability for quality control, testing, auditing, and compliance. Identity isn’t just the main driver for AI adoption; it’s at the center of the widespread interoperability that every organization demands.
Organizations have many choices when it comes to agent types: internal, external, off-the-shelf, and retail. While each has its own advantages, they all need to be trusted before they can be adopted. To gain that trust, they must be engineered with an identity-centric mindset. This is crucial for engineers who often don’t understand that they are accountable for security—and that identity needs to be at the heart of that.
The Result
When generative AI and agentic AI deployments center identity, they establish the guardrails and controls that build trust in AI and accelerate its adoption. Collaborating with partners like Okta and Amazon Web Services, who understand the importance of identity and have built it into their tools and services, can further speed implementation, empowering you to unlock the power of AI in exciting ways. When you are certain the data used to train AI models is appropriate and that the AI tools are acting responsibly, you can confidently use AI to innovate and create value for your business.
