Rubrik Launches ‘Agent Rewind’ to Undo AI Agent Errors Following Predibase Acquisition

Cybersecurity firm Rubrik has unveiled Agent Rewind, a new tool designed to help organisations identify, audit, and reverse errors made by autonomous AI agents. The launch comes shortly after Rubrik completed its acquisition of AI infrastructure provider Predibase.

Powered by Predibase’s technology, Agent Rewind provides enterprises with visibility into AI agents’ actions and the ability to roll back undesired changes to applications and data. The solution is targeted at sectors such as banking, financial services and insurance (BFSI) and healthcare, where regulatory compliance and transparency are paramount.

“In highly regulated sectors in India… maintaining transparency is critical,” said Satish Murthy, Chief Technology Officer for India and the Asia Pacific and Japan region at Rubrik. “The ability to safely rewind AI actions will give companies the confidence to experiment with generative and agentic AI without fear of irreversible mistakes.”

The launch addresses a growing concern over the reliability of AI agents, which, like human workers, can make operational mistakes. Industry research has highlighted issues ranging from technical malfunctions and legal missteps to the deletion of production databases. A recent study cited in Rubrik’s announcement found AI agents frequently becoming disoriented, taking incorrect shortcuts, and failing at multi-step tasks.

Anneka Gupta, Rubrik’s Chief Product Officer, said Agent Rewind combines Predibase’s AI capabilities with Rubrik’s recovery systems to “enable enterprises to embrace agentic AI confidently.” The tool offers detailed audit trails, context-enriched visibility into AI decision-making, and the ability to restore altered files, databases, configurations, or repositories through Rubrik’s Security Cloud.

The platform is designed to work with a range of AI development environments and services, including Salesforce’s Agentforce, Microsoft Copilot Studio, and Amazon Bedrock Agents, as well as custom-built AI agents.

Industry analysts say the product tackles an emerging challenge. “Agentic AI introduces the concept of ‘non-human error,’ and… organisations should explore solutions that allow them to correct potentially catastrophic mistakes,” said Johnny Yu, Research Manager at IDC.

Agent Rewind aims to go beyond current observability tools by not just revealing what actions an AI took, but also why those actions occurred and how to reverse them — a feature Rubrik says could help enterprises adopt AI systems more safely.