Arjuna Kodisinghe - stock.adobe.
How AI is changing data protection
There's no doubt that recent AI developments have affected data protection. IT leaders must not only stay on top of these changes but also prepare for what's next.
Data protection is a critical discipline for every organization, and the rise of AI has radically changed it. For better and for worse.
In some ways, the use of AI technology is increasingly enabling data protection, with new advanced capabilities that aim to improve outcomes. Gartner's June 2025 report on backup and data protection platforms found fewer than 25% of backup tools incorporated generative AI in 2025, projected to reach 90% by 2030.
Unfortunately, developments in agentic AI in particular have also created entirely new risks that organizations must now prepare for and mitigate. Gartner's 2025 cybersecurity trends report pointed out that the wave of GenAI, cloud and automation adoption improving workflows has driven a surge in machine accounts and credentials. Left unmanaged, those credentials can expand an organization's attack surface around backup systems and the data they protect.
AI adoption is on the rise, so IT leaders must be prepared for how it can both aid data protection efforts and expose a business to new risks.
Where AI already helps
Looking at the advanced features that AI brings, there are several key capabilities that aid data protection efforts, including the following:
- Anomaly detection. AI is particularly good at pattern matching. With AI, data protection platforms can compare each new snapshot against a behavioral baseline built from historical activity.
- Failure prediction. AI and machine learning models analyze historical and real-time data, including logs and performance information. Organizations can use that data to proactively predict potential failures.
- Backup schedule optimization. The system can analyze and understand usage and demand, then use that information to allocate data protection resources and time backups more efficiently.
- Data integrity. Ensuring data integrity has always been part of data protection. With AI, that functionality is enhanced for greater accuracy and performance.
- Recovery workflow automation. AI can draft a step-by-step recovery plan from an organization's infrastructure, rather than leaving that work to a manual process.
- Natural language. Instead of requiring specific training or knowledge, conversational AI is enabling data platforms to be easier to use with natural language queries.
Where AI causes problems
Like most new technologies, AI can also potentially lead to data protection problems in a variety of ways. Two major risks are destructive AI and data protection failures.
Destructive AI
Despite being fairly new, AI coding agents have already destroyed the data they were supposed to protect, backups included. There have been several publicly reported incidents to date.
In April 2026, an AI coding agent on Cursor, running Claude Opus 4.6, deleted PocketOS's production database and every backup in a single command, because the infrastructure provider stored both in the same volume.
In July 2025, a Replit AI agent deleted a live production database during an active code freeze, then told the company's founder that rollback was impossible. This turned out to be false, but the incident caused significant downtime and damaged the company's reputation.
Both organizations addressed these issues through platform fixes. Replit added automatic separation between development and production databases and a planning-only mode, and PocketOS added a delay before backups can be deleted.
Data protection failure
AI detection features inside data protection tools can also fail on their own.
In 2024, Veeam shipped an inline entropy engine in Backup & Replication v12.1 to catch ransomware by measuring how encrypted a data stream looked. However, in production, the machine learning model regularly flagged legitimate files as suspicious. Windows EFS files, ISOs and even browser cache data were caught in the false positives, according to reports on Veeam's own community forums.
Veeam has since rebuilt its detection around a different machine learning model that builds an individualized baseline for each dataset rather than applying one entropy threshold.
What vendors are building
Recent AI data protection announcements share a common thread. Vendors have expanded their data protection platforms with new AI capabilities, from classifying backup data for AI pipelines to querying it in plain language.
The table below tracks major data protection vendors that added significant new AI capabilities to their own platforms between July 2025 and July 2026.
Taken together, these announcements point in the same direction: Vendors are folding data classification, governance and recovery into single platforms instead of selling them as separate add-ons.
The acquisitions behind these launches confirm it. Cohesity's absorption of Veritas, Veeam's purchase of Securiti AI, and Commvault's acquisitions of Clumio and Appranix all moved governance and security capability directly into backup platforms rather than leaving it to standalone tools.
For buyers, that shift means evaluating a vendor's full AI stack rather than shopping for point products.
What's next?
It's still relatively early in the AI era, but there are indications already of where things are headed. The following are three clear trends that IT leaders should keep an eye on:
- Adoption is accelerating. Gartner's June 2025 report projects 35% of enterprises will run autonomous, agentic backup systems by 2030, up from less than 2% in 2025.
- Vendors are consolidating. Cohesity absorbed Veritas on December 10, 2024, and by January 2025 had folded Veritas' Alta Copilot into its own Gaia assistant rather than keeping it separate. Expect more of the AI copilots above to disappear the same way.
- New copilots mean new risk. Each one is a new credential with broad access to backup data, the kind of access that let coding agents wipe out PocketOS and Replit. IT leaders must be aware of the risks and limit access before letting an AI feature make a recovery decision on its own.
Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He has pulled Token Ring, configured NetWare and been known to compile his own Linux kernel. He consults with industry and media organizations on technology issues.
