Saviynt has launched Zuma, an identity security platform built to discover, govern and protect AI agents, large language models, AI-based applications and other non-human identities as organisations scale AI across the business.
The company said AI and non-human workers now outnumber humans by more than 100 to one, creating major risk that is currently unmanaged. Unlike dashboards that report risk after the fact, Zuma acts at runtime, evaluating an AI agent's intent, context and risk to stop unauthorised access in the moment.
Saviynt said it developed Zuma after working with large enterprises and finding that AI agents and non-human identities were multiplying across business functions, often without effective oversight. Legacy identity tools, it argued, cannot answer basic questions such as where all an organisation's AI agents are, who created them, what they are doing, who is overseeing them, and what control exists over the data and applications they reach.
Zuma answers those questions through three components in a single control plane. Zuma Insights continuously discovers AI agents, LLMs, AI-based applications and non-human identities, maps what they access and who owns them, and shows lifecycle changes on a timeline. Zuma Access applies protection at runtime. Zuma Governance extends established identity best practices, such as ownership and access review, to AI.
"AI is fundamentally changing the identity security equation," said Sachin Nayyar, chief executive of Saviynt. "Enterprises are no longer securing only human users. They are now responsible for AI agents and non-human identities that can decide and access systems at machine speed."
Saviynt positioned Zuma as a single, end-to-end platform spanning visibility, runtime protection and governance, in contrast to what it described as fragmented point tools.
Non-human identities, which include service accounts, machine identities, API keys and now AI agents, have become one of the fastest-growing categories in identity security. Their number and autonomy make traditional, human-centred review cycles hard to apply, which is the gap Saviynt is targeting with runtime, intent-aware controls that bring discovery, ownership and governance into one place.