Part 3: The Future of Autonomy – Securing Agentic Workflows
If LLMs are the "brain," AI Agents are the "hands." As vendors move toward securing agentic workflows, the stakes shift from data leaks to unauthorized actions.
In Parts 1 and 2, we explored how data protection vendors are using DSPM and CSPM to govern how AI "reads" data. But we are quickly moving into a new era: The Agentic Era.
Unlike a chatbot that simply summarizes a document, an AI Agent—powered by platforms like Google's Gemini or Microsoft's Copilot—can act. It can trigger a refund in a CRM, update a record in a production database, or move files between cloud buckets. While this level of automation is a productivity goldmine, it creates a massive "Identity Crisis" for primary data security.
The Shift from "Read" to "Write" Risk
The moment an AI is given "Agency," the security model must change. In a traditional setup, humans are the primary actors. In an agentic workflow, a Non-Human Identity (NHI)—the service account belonging to the agent—is the one making changes. Recent research from Rubrik Zero Labs highlights that non-human identities now outnumber human users by a staggering 82:1, creating a massive, unmanaged attack surface (
If that agent is compromised via Indirect Prompt Injection (where an agent reads a malicious email or file that contains hidden instructions), it could be tricked into using its "write" permissions to delete primary data or exfiltrate a database under the guise of a routine task.
How Data Protection Platforms are Evolving for Agents
To secure these autonomous workflows, data protection vendors are shifting from static backups to Active Governance:
Identity-Aware Data Resilience: Vendors like Rubrik and Veeam are now focusing on "Identity Resilience." This means ensuring that if an AI agent is compromised, its blast radius is contained. Rubrik’s new Agent Cloud initiative focuses on bringing these non-human identities into a "least-privilege" managed model (
).Source: CRN Asia AI-Specific Firewalls: Following its $1.725 billion acquisition of Securiti AI in late 2025, Veeam has integrated an "AI Firewall." This tool inspects the prompts and outputs of agents in real-time, preventing "Toxic Entitlements" where an agent might accidentally (or maliciously) be granted the power to modify sensitive primary data (
).Source: IT Vortex Real-Time Audit Trails for AI: Cohesity, through its Gaia platform and expanded partnership with Google Cloud, is creating a "Black Box Recorder" for AI. It allows enterprises to audit exactly which historical data an agent accessed and what actions it took, ensuring that "digital insiders" are as accountable as human ones (
).Source: IT Brief Asia
Conclusion: The New Guardian of the AI Frontier
The convergence is complete. Data protection vendors have moved from the "basement" of IT—handling tapes and backups—to the "front lines" of the AI revolution. By mastering DSPM, CSPM, and Identity Governance, these companies have become the essential guardrails for the modern enterprise.
In 2026 and beyond, the most successful companies won't just be the ones with the fastest AI; they’ll be the ones who can prove their AI is safe, because they’ve secured the data and identities that power it.
Comments
Post a Comment