HYCU made aiR Graph generally available on Thursday, and the price is the headline: free, for anyone, customer or not.
What it does. aiR Graph connects read-only to Microsoft Entra ID, including Entra Agent ID where it’s enabled, and to Okta, including Okta AI Agents where the customer subscribes. The first scan maps applications and flags copilots and agents. For each agent it shows OAuth scopes, which data sources it can read or write, whether it has organization-wide access, whether it can act unattended, how many users can invoke it, and whether anyone is recorded as its owner. Severity scales with what an agent can reach and change. Then comes the pitch: it shows which of those applications have no independent backup, which HYCU R-Cloud would provide.
One design choice is worth copying. HYCU says it kept language models out of the risk assessment, so every finding comes from an explicit rule applied to identity-provider data, and the same tenant gives the same answer twice. For anything headed to an auditor, that’s the right call.
Why a backup vendor is doing this. Recovery is the one agent control that doesn’t depend on predicting what the agent will do. Guardrails, policies and runtime blocks all try to catch a bad action before it happens. A backup assumes some will get through and asks whether you can roll back. When an agent with write access deletes or rewrites records in Salesforce, GitHub or Confluence, the SaaS platform does exactly what it was told. If there’s no independent copy, that’s the end of it.
Rubrik made this argument first. It introduced Rubrik Agent Cloud with a waitlist in October 2025 and reached general availability in February with agent discovery, policy enforcement on prompts and tool calls, and Agent Rewind for rolling back agent-driven changes. A version for Claude Code followed in June. HYCU’s move is cheaper and narrower: the map is free, and the paid product is the backup.
Our POV. This is a lead-generation tool, and that’s fine. It’s also a signal about where agent inventory is heading. The identity providers already register agents through Entra Agent ID and Okta AI Agents. A backup vendor can now read that registry and hand you the result for nothing. That squeezes startups whose main product is “find all your agents.” Discovery on its own is becoming a feature. The money is in what comes after: block, govern or recover.
The gap in the map. An inventory built from the identity provider only sees agents that authenticate through it. Coding agents on developer laptops, scripts holding long-lived API keys and vendor copilots calling your APIs with their own credentials may never show up. On the same day, Arcjet shipped an endpoint agent specifically to find unmanaged coding agents. If you run aiR Graph, treat the count as a floor, not a census.
What we’d do. Run it. It’s read-only and free, and the output is a useful gap list. Check exactly which scopes the connector requests before you grant them, and ask where the scan results are stored, because a full view of your identity provider is sensitive on its own. Compare its agent count with what your CASB, endpoint and API gateway logs show. The difference is your shadow-agent problem. Then sort the apps agents can write to by whether they have a backup outside the SaaS vendor. That list is where an agent mistake turns into permanent loss.
The market read. Data-protection vendors have spent two years repositioning as cyber recovery. Agents give them a new way in: you can’t stop every bad action, so you’d better be able to undo it. We’d expect other backup vendors to follow with free agent scans of their own. Once several vendors give away the same map, the map stops being the product, and the undo button is what gets bought.