Enterprises are buying an identity layer for their agents

The gap: agents already have production access

The numbers in Saviynt's 2026 Global Survey on SoD Risk Management are worth sitting with: 38% of deep-dive respondents already have AI-driven actions reaching production SAP transactions. And of 21 organizations with agents in production, only 2 have fully enforced separation-of-duties guardrails. This is a vendor survey, so treat the figures as directional — but the direction is unmistakable: agents are touching production systems faster than the controls meant to restrain them. m:…6215

Saviynt's Simon Moffatt frames the problem as identity, not tooling: agentic identities combine non-human identity risks with human-centric IAM risks and new autonomous risks. Traditional IAM controls were built for predictable workloads or accountable humans, and agentic identities often carry excessive permissions with broad system access, similar to humans. That is his opinion, from an essay on Saviynt's blog — but it names the gap precisely: the agent inherits the human's access without the human's accountability. He also warns that agentic identities are logically capable of bypassing simple policy-based guardrails, a point with sharp teeth for anyone relying on static policy checks. m:…5868

The product layer: identity for non-humans

Saviynt's answer is Zuma for Apps, introduced this week to govern AI agents and non-human identities inside business-critical applications. It maps the full chain: the humans who can invoke an agent, the authority delegated to it, the sub-agents and tools it can call, and the fine-grained entitlements it exercises inside target applications. It's in private preview for now — a launch, not yet a product with a customer list. m:…6215

Defakto took the standards route. The company announced that "Defakto Security joined the OpenID Foundation as a Sustaining Corporate Member." Its stated aim is to eradicate long-lived credentials for non-human identities. CEO Danny Oliveri says AI is accelerating the trend of non-human identities needing access. At one large retailer, the company says more than 12,000 applications already rely on Defakto identities. m:…6592

Governance as the commercial wrapper

Workato is selling the same layer from the automation side. The company frames AI governance plainly: "AI governance controls what an autonomous agent is allowed to do and provides a record of its actions." To back the scale claim, Workato says it has processed more than 1.1 billion enterprise AI actions as of September 2026, and that it achieved ISO/IEC 42001 certification in March 2026. Those are the company's own numbers — but the framing is the interesting part: governance as the answer to what the agent may attempt, on whose behalf, and what happened. m:…9212

The technical shape of that governance is also worth noting: Workato says it runs four gateways — Model, MCP, Agent, and API — that share one policy and audit plane. That's the vendor's bet on where the category converges: many entry points, one place where identity and policy are decided. m:…9212

Put the three together and a pattern emerges: identity has become the commercial wrapper for the agent era. This is my read, not a reported fact — the models are commoditizing, the orchestration is cheap, and what enterprises will actually pay for is the permission-and-record layer in between. Every announcement here comes from the company selling it. But the fact that three of them landed on the same day suggests a category is forming, not just marketing. m:…6215 m:…6592 m:…9212