Veeam on Tuesday introduced Agent Commander, a new set of features aimed at detecting and mitigating AI risk.
Integrating AI is a prominent initiative for many enterprises today. In particular, integrating agentic AI capabilities is a priority given that agents can be trained to autonomously make decisions, take actions and execute certain business processes at machine speed.
However, agentic AI can propagate mistakes at that same level, according to Johnny Yu, an analyst at IDC.
"While part of the point of using AI agents is giving them some autonomy, organizations should have guardrails in place to prevent abuse of that autonomy," he said.
Safe data usage is one of the biggest hurdles to large-scale agentic AI deployment, and customers expanding agentic AI use must that ensure agents can't access personally identifiable information, trade secrets and other sensitive data, Yu continued. In particular, he noted that misconfiguration of AI agents, including granting them too much access to sensitive data, is a major risk associated with the push to implement agentic AI across an organization.
Agent Commander, which will be available in private preview at the end of March as part of the Veeam Data Cloud for Microsoft 365, is designed to counter potential AI risks with capabilities that detect AI risk, protect AI systems and undo AI actions.
Agent Commander detects AI risk with context, identifying shadow AI, sensitive data exposure and risky agent behavior. Protection includes enforcing granular, real-time controls across environments. And undoing refers to performing precise, context-aware recoveries to restore data that agents have erroneously deleted or changed.
While part of the point of using AI agents is giving them some autonomy, organizations should have guardrails in place to prevent abuse of that autonomy.
Johnny YuAnalyst, IDC
The protections were designed to help customers more confidently pursue agentic AI initiatives, according to Rick Vanover, VP of product strategy at Veeam.
"Leadership of organizations across the world are pushing their teams to drive AI solutions, and this is just a risk reduction technique," he said.
Added Securiti
Agent Commander is the first integration to come from Veeam's $1.7 billion acquisition of Securiti AI last December.
Key to Agent Commander is Securiti AI's Data Command Graph, according to Vanover. The Data Command Graph is a real time AI-powered intelligence engine that maps an organization's data across different environments -- including production and backup data – to provide customers with a visualization of the AI agents they've deployed and what those agents can access.
By providing customers with an overhead view of what AI agents can access, Data Command Graph helps rein in rogue agents and ensures that agents are working with only necessary, high-quality data to perform their functions.
"That is the heart and soul of why we acquired Security AI, in the sense of good data will fuel good AI," Vanover said.
Risks associated with agentic AI are a relatively new concern when it comes to data protection. However, in addition to Veeam's Agent Commander, there are other tools targeting the problem, according to Yu.
Predibase, recently acquired by Rubrik, enables organizations to see what AI agents they've deployed and what they can access. Like Agent Commander, Rubrik's Agent Rewind lets users undo AI agent actions and gives context and reasoning on how an agent arrived at its decision.
Beyond targeted tools, there are ways to rein in agentic AI risk, according to Yu. However, they are not as effective, he added.
"Organizations without such tools usually lower agentic AI risk by simply not allowing their agents to do anything risky," Yu said.
For example, they can limit agents' access to datasets or prevent them from taking major actions without human approval.
"The drawback to this is that they are limiting the potential -- and arguably, the purpose -- of AI agents to correctly and consistently take actions at machine speed," Yu said.
Using agentic AI is a delicate balance between letting agents do their job and protecting enterprise data from risk, but Agent Commander is a step in the right direction, according to Yu.
"Veeam Agent Commander directly addresses the challenge of accountability and auditability of AI agents' actions," he said.
Looking ahead
Agent Commander is just the first set of tools Veeam is releasing that integrate capabilities the vendor acquired when it bought Securiti AI. However, more releases are on the horizon, according to Vanover.
Currently, Agent Commander is only available in Veeam Data Cloud, the vendor's first-party backup-as-a-service for Microsoft 365. Beyond Agent Commander, Vanover said that he expects additional features to go to market in the future.
Meanwhile, because Agent Commander directly addresses an important market need, Yu noted it was not surprising that Agent Commander was the first release to come from the acquisition. However, there are several other capabilities inherited in the Securiti AI acquisition that would be valuable to integrate into Veeam's platform, according to Yu.
"Securiti AI's bread and butter is identifying ROT -- redundant, obsolete, or trivial -- data, and I expect some focus on that capability in a future release," he said.
Erin Sullivan is a senior site editor at Informa TechTarget covering data technologies.
