Three security vendors announced products on November 10 aimed at the same new problem: AI agents that act inside company systems with real permissions. Each tackles a different moment. Cyera wants to judge an agent’s intent. Obsidian wants to approve or block each action before it runs. Druva wants to restore an agent when something goes wrong anyway.
Cyera: intent, endpoints and its own agents
At its datasecAI 2026 event, Cyera added two products to its Cyera Omni platform. Cyera Endpoint pushes policy enforcement onto devices, where local AI apps and agents now run. It discovers them and monitors file access, command execution, scripts and tool calls, linking each action to the identity behind it and the data involved. Cy Agents are Cyera’s own specialised security agents that share business context and take on investigation and risk work under human direction.
Cyera’s argument is that at agent speed, telling legitimate work from risky behaviour depends on intent: what the agent is trying to do and whether it fits a business purpose. The company has the money to pursue it. It added a $400 million Series G extension from Goldman Sachs Alternatives in September and has bought several startups, including Oasis Security for non-human identity management.
Obsidian: a decision engine for every action
Obsidian Security announced what it calls the first context-aware runtime decision engine for agent actions across third-party applications. When an agent tries something, the engine weighs who the agent acts for, what the app permits and how sensitive the data is. A dropped table, a mass delete or customer data sent outside can be held for approval or blocked while the agent’s other work continues.
Policies are written once and applied everywhere agents act: coding tools such as Codex, MCP connections and existing gateways. A Policy Decision API lets other infrastructure send agent actions to Obsidian and enforce the verdict. Cloudflare is one integration partner. Every decision is logged for auditors.
Druva: backup for the agents themselves
Druva introduced Agent Recovery, starting with backup and restore for Microsoft Copilot Studio agents. As more business processes run through agents, a deleted or broken agent becomes an outage. Druva keeps a version history for each protected agent and can restore a single agent and its configuration without rolling back the whole Power Platform environment. An integration with Microsoft’s Azure SRE Agent arrives at the end of November. Druva will demonstrate the feature at Microsoft Ignite, November 17 to 20.
The pattern
Put next to Bay, the Israeli startup that came out of stealth the same day with $50 million to vet agent actions (covered on technologies.org), and Vanta’s purchase of Pillar Security a day earlier, the market is filling out fast. Discovery, runtime enforcement, data context and recovery are all being built at once. Buyers will soon have to decide whether they want one platform that does all four or best-of-breed tools for each.