A New Identity: Agentic AI boom risks busting IAM norms

Artificial intelligence isn’t just transforming identity security, it’s reinventing it faster than most teams can react. Agentic AI, a class of autonomous, collaborative software agents, has emerged as the latest disruptive force shaping enterprise technology today.

Major research firms like Aragon are calling out the marriage of agentic identity and security platforms (AISPs) as a core strategic trend for enterprise security. But while these developments promise stunning new capabilities, they also introduce urgent challenges for identity and access management (IAM).

Agentic AI is beginning to power everything from dynamic access controls to automated software engineering. But the explosion of autonomous behavior is also generating an overwhelming number of machine identities, and with them, a tangle of authorization sprawl and over-permissioning that security teams are struggling to keep pace with and contain.

Standards bodies and industry-specific regulators must step in now. While it’s well understood that standards-setting moves slowly, the velocity of AI innovation and the risk it introduces, demands accelerated IAM security guidance. This is particularly true in highly regulated sectors such as healthcare and finance. Here, security leaders can’t afford to wait for consensus to form years down the road.

A full 78% of CISOs admit that AI-powered cyber threats are already having a significant impact on their enterprises, according to Darktrace’s 2025 report, "The State of AI Cybersecurity." And the risks are real. AI agents can be manipulated through prompt injection, corrupted by poisoned communication with other agents, or even used to silently alter system configurations. Researchers at Anthropic recently demonstrated how AI agents could be co-opted to exfiltrate sensitive data.

The core problem? Identity systems weren’t designed for this. Traditional IAM frameworks assumed humans, devices, or static service accounts. They weren’t built to govern semi-autonomous, ever-changing agents that replicate, delegate tasks, and act on behalf of other entities. Security teams are now being overwhelmed by a surge in unmanaged or misunderstood identities, often granted excessive privileges that linger long past their usefulness.

While agentic AI may offer new levels of efficiency and scale, it demands a rethinking of IAM from the ground up. Identity is still the cornerstone of enterprise security, but its scope is expanding rapidly, pulled in all directions by intelligent automation. Without a framework for governing these agents that track who they are, what they can access, when their rights expire, we risk chaos through complexity.

There is hope. Emerging open standards like Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication offer early blueprints for how these agents can discover, share information, and collaborate securely. But these standards must be backed by policy, tooling, and cross-industry cooperation. And that cooperation must happen fast.

MCP and A2A communication are emerging open standards designed to help AI agents work together more effectively. MCP allows data owners to securely expose information to AI agents in a controlled, structured way. A2A enables different agents — possibly from different vendors or platforms — to discover each other, share information, and delegate tasks. Together, these protocols lay the foundation for secure, scalable collaboration between AI systems in complex environments.

AI technologies may have once seemed like a distant bell. But today, they’re sounding loud and clear — and identity is what’s at stake.