The data and analytics landscape is rapidly evolving to meet the demands of agentic AI and intelligent automation. By 2026 and beyond, enterprises will shift from centralized, gatekeeper-driven models toward federated, real-time data architectures that prioritize agility, governance, and trust, according to the 2026 IDC FutureScape Report: Worldwide Data & Analytics 2026 Predictions1.
The same report states that by 2027, growth of AI agents will push 50% of CIOs to restructure and automate identity and data access and authorization management to reduce misuse and leakage as part of a zero trust architecture
In the near term, AI agents will be essential for scaling cybersecurity, triaging alerts, and quantifying risk in business terms — because human capacity alone cannot keep pace with escalating threat volumes. At the same time, governments are asserting sovereignty over data and AI systems, embedding compliance and residency requirements into the foundation of enterprise architectures, according to the 2026 IDC FutureScape Report: Worldwide Security & Trust 2026 Predictions2.
“Data is the foundation of good AI – yet it’s also the biggest liability to AI initiatives,” said Jennifer Glenn, IDC Research Director, Information and Data Security. “Protecting sensitive and confidential data in an AI world requires more than preventing exfiltration. Efficient and effective AI initiatives require strong protection and control around the data itself, allowing the data to remain secure as it is moved, shared and used.”
Executives, CISOs, CIOs, and Boards of Directors are all wrestling with the same questions as they look to the year ahead:
- How do we monetize GenAI and Agentic AI responsibly?
- How do we maintain data security, privacy, and regulatory alignment as AI accelerates?
- What’s next, RAG (retrieval-augmented generation), post-quantum threats, compliance challenges, advanced attacks, and are we ready?
- How do we invest in 2026 data priorities without locking into brittle architectures we will regret?
Below is Paperclip’s assessment of existing challenges around AI security, including example use cases, risks, and potential solutions.
The AI Data Exposure Challenge
Traditional encryption approaches are failing in the AI era—an explosion in new data usage; traditional database encryption was never designed for. The conventional encrypt-decrypt-re-encrypt cycle is not only impractical; it also creates dangerous vulnerability windows at precisely the moments when data is most valuable—and most at risk.
Every time a model, application, or AI agent needs to access sensitive information, traditional approaches require decryption, leaving data exposed in memory. Agentic AI compounds this problem exponentially: unlike human users who make occasional data requests, AI agents generate thousands of queries per second, each one an exposure point.
Memory-based attacks targeting decrypted sensitive data represent a rapidly growing threat vector. When data must be decrypted for every operation, the attack surface expands with every AI interaction.
What Is Encryption-in-Use?
Encryption-in-use, also called Advanced Encryption or Searchable Encryption, represents a fundamental shift in how organizations protect data. Unlike encryption-at-rest (which protects stored data) or encryption-in-transit (which protects data moving between systems), encryption-in-use protects data during active processing or production.
The key capability: operations can be performed on encrypted data without ever exposing plaintext—data always remains encrypted. Searchable encryption technologies enable search, filter, and match operations while data remains fully encrypted. This eliminates the vulnerability windows that legacy database encryption usage creates.
For AI workflows, this means:
- RAG pipelines can query encrypted knowledge bases
- Analytics can run on always encrypted datasets
- AI agents can access the information they need without exposing sensitive values
The implications for AI security are profound. When data never needs to be decrypted for routine operations, the traditional tradeoff between utility and security dissolves.
GenAI Security
Use Cases and Examples
- Real-Time, Federated Data Access for AI and Agents
Use case:
You are rolling out GenAI copilots and agentic AI services across multiple business units. They need consistent, real-time access to customer, transaction, policy, and claims data scattered across clouds, regions, and legacy platforms.
The risk: If each new AI agent or microservice calls back to plaintext databases, data lakes, and caches, you quietly multiply your attack surface. One compromised service account or misconfigured API will leak sensitive data at scale. “Federated” becomes another word for “many places to exfiltrate from.”
What’s needed: Organizations require SAFE advanced encryption so that security controls scale and work across domains—keeping sensitive fields encrypted across operational systems, clouds, and business units while still supporting search, filter, and match operations. AI, analytics, and RAG pipelines must work with encrypted values instead of raw PII, PHI, or confidential data. Central governance with local control becomes essential: encryption policies managed centrally, while data stays in the domains where it belongs.
- Unified HTAP Workloads: Converging Analytics and Transactions
Use case: You consolidate operational and analytical workloads into a unified or HTAP platform to support real-time personalization, fraud detection, and AI-driven decisioning from a single data plane.
The risk: If HTAP platforms run their most important queries and AI workloads on decrypted datasets, they become exceptionally attractive breach targets. A compromise exposes not only today’s transactions but years of customer history.
What’s needed: A way to make HTAP and unified platforms always-encrypted by default without breaking performance. SAFE encrypts data at the field level and supports search, filter, and correlation over encrypted values. That lets you keep a unified view of customers and transactions while preserving data security, privacy, and auditability for regulators and boards.
- Data Exchanges and Cross-Enterprise Collaboration
Use case: You participate in data exchanges, clean rooms, or joint analytics programs with partners, reinsurers, or ecosystem platforms to build better risk models, pricing, or customer experiences.
The risk: Cross-enterprise data collaboration accelerates product innovation but, at the same time, increases cross-enterprise breach impact. The more raw data leaves organizational control, the harder it is to contain. As soon as raw data leaves your environment, you have limited control over how it is used, copied, or protected. Even with contractual controls, a partner breach can become your headline.
What’s needed: SAFE provides privacy-enhancing cryptography (PEC) that enables partners and clean rooms to perform joins and analysis on encrypted identifiers rather than exchanging raw PII or PHI. “Never in the clear” policies must be enforceable around the most sensitive data fields while still enabling high-value use cases.
- Cold Storage Becomes Strategic: AI-Ready Data Archives
Use case: You want to apply GenAI and advanced analytics to decades of policy, claims, customer, or transaction archives to uncover hidden patterns and drive new revenue or risk insights.
The risk: If legacy data archives are rehydrated and decrypted to feed AI, old liabilities become new exposure. A breach of reactivated archives can be worse than a modern production compromise.
What’s needed: A strategy that treats data archives as strategic yet always protected assets. SAFE lets you store and query archives as encrypted, searchable datasets so GenAI and RAG can find patterns without exposing raw content. That supports data resilience and compliance while still monetizing historical data. When regulators or auditors ask why “cold” data archives are being reused, organizations must demonstrate that sensitive fields never leave encrypted form—even when queried or referenced in AI prompts.
- Data Governance and “Who Saw What”
Use case: You are building enterprise-wide observability and data governance to satisfy regulators, internal audit, and your board. You need to prove not just where data moved, but how it was protected and who actually accessed sensitive fields.
The risk: If encryption is implemented as a patchwork of point solutions, no one can answer basic questions such as “Was this field encrypted here?” or “Which identities accessed that data last quarter?” You end up with strong slides and weak evidence.
What’s needed: A centralized SAFE encryption-in-use platform with cryptographic audit trails that track operations including key usage and field-level access patterns. This enables answering “who saw what” across AI pipelines, analytics jobs, and agentic workflows. Data governance and lineage tools must be enriched with encryption metadata: which fields were encrypted, when they were searched, how they were returned. Encryption must become an operational control that can be measured, reported, and improved—not a black box.
- Agentic AI, Zero Trust, and Non-Human Identities
Use case: You deploy agentic AI that can take actions on systems: open tickets, update records, submit requests, or recommend decisions that humans approve or override.
The risk: Agents quickly outnumber humans. If those agents can pull broad datasets in plaintext, any compromise or misconfiguration can lead to mass data exposure. Traditional role-based access control is not designed for thousands of non-human identities (NHI) with dynamic privileges.
What’s needed: A zero-trust approach to data, not just to networks and identities. SAFE enforces field-level encryption and policy, so agents only ever see what they must see, only what they’re uniquely authorized to see. Different agents can be bound to different keys and scopes, reducing blast radius and aligning AI security with your zero-trust strategy.
- AI-Driven SOC, Detection, and Automated Response
Use case: You use AI to triage alerts, enrich incidents, and propose or execute response actions across endpoints, cloud, and applications.
The risk: SOC tools ingest and store highly sensitive telemetry: payloads, user identifiers, transaction details, and notes from investigations. If all of that is in plaintext, a breach in your logging or case management platform becomes just as damaging as a breach of core applications.
What’s needed: Data security controls inside security operations themselves. SAFE keeps sensitive values in logs, alerts, and case notes encrypted, while letting SOC tools work with what they need. For deeper investigations, SAFE supports scoped, audited decryption. Detection stays powerful, but the data underneath remains resilient if tools or users are compromised.
- Sovereign AI, Data Sovereignty, and Digital Trust
Use case: You operate in jurisdictions that require in-country processing, sovereign AI stacks, and strict data sovereignty for citizen or customer data.
The risk: Simply placing infrastructure within a given country, or jurisdiction is not enough. If sensitive data is freely decrypted in memory, caches, or AI pipelines, you may meet residency requirements on paper but still fail privacy and security expectations in practice.
What’s needed: A sovereign encryption boundary that travels with the data. SAFE keeps data encrypted across regions and environments and only decrypts it inside tightly controlled services. GenAI, RAG, and analytics can operate on encrypted content while you maintain data sovereignty, privacy, and trust with regulators and citizens.
- Quantum Risk, Post-Quantum Readiness, and Long-Lived Data
Use case: You are starting to plan for post-quantum cryptography and the risk that adversaries might harvest encrypted data today to decrypt later.
The risk: Long-lived data archives, high-value identities, and sensitive records are attractive targets for harvest-now, decrypt-later strategies. If encryption is implemented differently in every application, migrating to quantum-safe algorithms becomes slow, expensive, and inconsistent.
What’s needed: A crypto-agile architecture for your most critical data sets. SAFE centralizes already post-quantum resistant advanced encryption-in-use, and crypto-agility, so you only evolve algorithms and key strategies once, instead of refactoring dozens of bespoke implementations. That advances post-quantum readiness and enhances the resilience of data that must remain protected for years or decades.
- Private Cloud Compute, Cloud LLMs, and Hybrid AI
Use case: You adopt Private Cloud Compute, confidential computing, or similar models to run LLMs and GenAI workloads on sensitive data while using public cloud scale.
The risk: Even in protected runtime environments, feeding large volumes of raw training data, prompts, and context in plaintext increases exposure. A configuration error, insider threat, or future vulnerability in the runtime stack can expose the very data you tried to safeguard.
What’s needed: Consistent, advanced encryption controls that span on-prem, cloud, and hybrid AI. SAFE encrypts sensitive portions of knowledge bases and contextual data feeding LLMs and agentic AI. Prompts and retrieved context can stay encrypted under SAFE control, with decryption happening only where justified and governed. You gain the benefits of GenAI and agentic AI without sacrificing data security, privacy, or sovereignty.
How Paperclip SAFE Addresses These Challenges
The challenges outlined above share a common thread: traditional, legacy encryption approaches cannot support the security demands of the AI era; it’s just now how they were designed. Organizations need a new, advanced approach—one that eliminates the decrypt-process-re-encrypt cycle entirely.
Paperclip SAFE® Platform is purpose-built for this moment. As a privacy-enhancing cryptography (PEC) solution, SAFE combines encryption-at-rest and in-transit with encryption-in-use through searchable symmetric encryption combined with Paperclip’s patented technology and processes.
What makes SAFE different:
SAFE enables workloads to operate directly on encrypted data. Whether performing CRUD operations, running analytics, or powering RAG pipelines, sensitive information remains encrypted throughout. AI agents are treated as authenticated non-human identities with fully controlled, zero-trust access.
Addressing the federated data challenge: SAFE keeps sensitive fields encrypted across operational systems, clouds, and domains while supporting search, filter, and match operations. Agents and applications query encrypted indices without seeing underlying values.
Enabling unified platforms: SAFE encrypts at the database layer, so the same encrypted values support both operational and analytical workloads—no decryption detours required.
Supporting collaboration: SAFE acts as a PEC layer where partners and clean rooms perform analysis on encrypted identifiers, enforcing “never in the clear” policies.
Protecting archives: SAFE secures archived content at the field level, enabling AI to mine historical data without plaintext exposure. Searchable encrypted history supports RAG workflows while maintaining zero-trust guardrails.
Delivering auditability: SAFE creates cryptographic audit trails of all operations—key usage, field-level access, policy enforcement—feeding governance and compliance requirements.
Preparing for quantum threats: SAFE’s crypto-agile architecture supports post-quantum algorithm adoption through a centralized upgrade path, with patented data shredding providing additional protection against harvest-now-decrypt-later attacks.
For CISOs, CIOs, CEOs, and Boards of Directors, this translates into practical 2026 priorities:
Move from “encrypt, decrypt, re-encrypt” to “encrypt and use.” Adopt advanced encryption and PEC to protect data in active processing across AI and analytics workflows.
Treat AI security and data governance as operational disciplines. These are core business requirements, not side projects. Cryptographic audit trails must support regulatory compliance.
Build centralized post-quantum readiness plans. Choose crypto-agile architecture over siloed system-by-system migration. Protect long-lived archives from harvest-now-decrypt-later attacks.
Increase data resilience for AI reuse. Strategic datasets will power AI for years to come. Always-encrypted archives enable reuse without reexposure.
Paperclip SAFE is designed to support these priorities—serving as an encryption and privacy fabric that keeps critical data always encrypted (including in use), enables AI to operate on protected information, and feeds governance with verifiable cryptographic evidence.
Frequently Asked Questions (FAQ)
What is encryption-in-use and how does it differ from encryption-at-rest?
Encryption-in-use is an advanced encryption specifically designed to protect data during active processing (production), while encryption-at-rest only protects stored, static data. Traditional encryption requires decryption before use, creating vulnerability windows. Encryption-in-use technologies like searchable encryption allow operations on encrypted data without exposing plaintext, enabling secure AI workflows.
Why is agentic AI a security concern for enterprises?
Agentic AI systems operate autonomously and require real-time access to sensitive data across federated systems. Unlike human users, AI agents generate thousands of data requests per second, dramatically expanding the attack surface. Without encryption-in-use, every AI agent interaction becomes a potential data exposure point.
What is post-quantum encryption and when should organizations start preparing?
Post-quantum encryption uses algorithms resistant to attacks from future quantum computers. Organizations should begin preparation now because adversaries are already harvesting encrypted data today to decrypt once quantum computers become available (harvest-now-decrypt-later attacks). Long-lived data like medical records, financial archives, and classified information require immediate post-quantum protection.
How does privacy-enhancing cryptography (PEC) support data governance and AI compliance?
PEC technologies like searchable encryption create cryptographic audit trails showing exactly which data was accessed, when, and by whom—including AI agents. This supports data governance frameworks, regulatory compliance (GDPR, HIPAA, CCPA), and AI Bills of Materials (AI BOMs) by providing verifiable evidence of data protection throughout the AI lifecycle.
Can organizations use encryption-in-use with existing AI and analytics platforms?
Yes. Modern encryption-in-use solutions integrate at the API and database layer, working alongside existing HTAP platforms, RAG pipelines, and data lakes. Organizations can implement searchable encryption without replacing their AI infrastructure, protecting sensitive fields while maintaining analytical and operational performance.
About Paperclip
Paperclip: Advanced Encryption for the AI Era
Paperclip specializes in encryption-in-use technology that protects sensitive data during active AI processing, analytics, and operational workflows. Our SAFE® Platform combines encryption-at-rest, in-transit, and in-use with searchable encryption capabilities, enabling organizations to deploy agentic AI, GenAI, and advanced analytics without exposing critical data to next-generation threats.
As a privacy-enhancing cryptography (PEC) solution with post-quantum readiness, SAFE helps CISOs and data governance leaders implement encryption strategies that turn AI security from a roadblock into a competitive advantage.
Ready to build always-encrypted AI workflows? Let’s start a conversation about how encryption-in-use can support your 2026 data priorities. Contact Paperclip to discuss your AI security strategy.
Sources:
- 2026 IDC FutureScape Report: Worldwide Data & Analytics 2026 Predictions – Oct 2025 – Doc # US53860225
- 2026 IDC FutureScape Report: Worldwide Security & Trust 2026 Predictions – 4-Nov-2025 – Doc# US53858325