Why Identity Is the Foundation of Agentic AI Security
Agents have arrived as the next wave of AI usage, especially when it comes to extending the utility of large language models (LLMs).
Agents also expand AI’s blast radius when it comes to security. Take, for example, agents’ nondeterministic nature. They won’t trace the same path through the network every time, and they won’t necessarily follow predictable courses of action even for simple tasks.
They’re also sent forth to access infrastructure and data—and they’re automated. The combination makes them juicy targets for hackers. But even if we assume no agents are malicious, it’s best to view them like mischievous toddlers. You don’t want them to experiment too far.
Futuriom, in collaboration with Teleport, has prepared this Leadership Brief to explain the sea change in the concept of identity. Applying the healthy paranoia that’s always useful in security discussions, we can think of agents as a relentless, unpredictable swarm bombarding infrastructure. Clearly, we want to limit the number of ways these agents can achieve bad outcomes.
Identity Needs a New Philosophy
It’s a multifaceted issue, but there’s a clear starting point. An agentic AI security plan must be built upon a rigorous approach toward identity, one that applies across all systems.
This is not to diminish other types of security, not even the existing identity access management (IAM) or privileged access management (PAM) platforms. We’re talking less about technology and more about the approach. And identity is the focus because, for nearly all organizations, it’s become fragmented, leaving vulnerabilities that have gone unnoticed.
A classic example is a corporate merger. Two companies might have different identity systems, as well as different public cloud services and different data architectures—and those differences will likely stay in place. There might be a plan to merge them eventually, but let’s be honest: Do you think that happens every time?
That’s one way in which identity fragments. Additionally, many if not most users have multiple identities within an organization. Some platforms or SaaS services might have incompatible identity mechanisms, causing people to have multiple logins. The easy way around that is to just live with the difference. People can be assigned multiple identities within a given system, too—maybe one for everyday work and admin “superuser” access for fixing problems.
Then there’s the issue of overprivileged access. IT admins, executives, team leaders—many types of personalities will end up with more permissive access than they should, sometimes for good reasons, sometimes just because they can. At older organizations, especially, it’s possible that many of these identities were never disabled, even after the person in question left the company. These invisible back doors just linger as potential targets for hackers.
Agents Find the Gaps
Another possibility to consider is that agents might stumble onto flaws in identity or permissions. Agents behave stochastically; one agent’s behavior might vary from one iteration of a task to another. Figuratively speaking, agents wander. When multiplied to agentic scale and speed, that means there’s a decent probability of an agent eventually doing something novel that turns out to breach security or violate policy.
Defending against those attacks (or mistakes) is a tall order, and we’re not claiming the answer is simple. But no strategy, framework, or platform is going to be sufficient if it’s built on a shaky foundation. That’s why now is the time to take a refreshed, rigorous approach to identity.
Key things you will learn in this Leadership Brief:
- Why identity needs a higher profile in your cybersecurity architecture.
- Where the dangers lie in existing identity frameworks, and why agents will almost certainly uncover them.
- How MCP and similar protocols magnify the potential trouble that agents can cause.
- Why scale, speed, and automation are problems but also part of the solution.
- Why organizations need to adopt a life-cycle approach toward identity and privileges.
