It’s 8:12 a.m. on a Monday. Your AI pilot is queued for the board this week. Security has flagged three issues. Finance wants the true total cost of ownership. Your data team is quietly exporting fields into a staging area because the application can’t run its lookups without plaintext. None of this is unusual. It’s the modern enterprise, where ambition outruns guardrails and where one question hangs over everything: how do we protect what matters without slowing down the business? IDC’s latest C‑suite snapshot highlights the same tension: protect sensitive data while enabling rapid AI‑driven objectives, with complexity and cost close behind and compliance cited by a much smaller share of respondents.
This is the design center of Paperclip SAFE® (Searchable and Fast Encryption). SAFE keeps high‑value data encrypted not just at rest and in transit, but during searches and processing—what’s known as encryption-in-use. Applications and AI agents can match, join, and retrieve what they’re allowed to use without exposing raw values in app tiers, caches, logs, or data lakes. Encrypt at source, govern at use, keep moving.
The wider market picture reinforces this need for enablement‑grade security. Cyber risk is becoming systemic as interdependence and complexity rise, according to the World Economic Forum’s Global Cybersecurity Outlook 2025. Spending tracks that urgency: end‑user investment in information security is projected to reach about $213 billion in 2025, and AI spending is set to approach $1.5 trillion. Meanwhile, the breach data hasn’t gone gentle: credential abuse and human‑driven errors remain central to incidents, per the Verizon 2025 Data Breach Investigations Report. And although the global average cost of a breach has dipped thanks to faster identification and containment, the U.S. average climbed to a record high, according to IBM’s 2025 Cost of a Data Breach. Translation: you need controls that shrink exposure even when identity or process fails, and that don’t block the AI gains your board expects.
Why Traditional Data Protection Approaches Fail for Modern Enterprises
Most security programs still assume that once data reaches an application, it must be decrypted to be useful. That assumption creates a sprawl of exceptions, masking scripts, DLP rules, and fragile compensating controls. It also guarantees that, somewhere, sensitive fields appear in plaintext where attackers, misconfigurations, or over‑eager AI pipelines can grab them. The Verizon DBIR has been consistent on this point for years: credentials are stolen, people make mistakes, and complex estates leak in unexpected ways.
Not sure which option is right for you? see SAFE pricing for your organization Choose the SAFE deployment model that fits your security requirements.
Encryption‑in‑use inverts that logic. With SAFE, sensitive values remain encrypted while your authorized workflows perform policy‑aware queries and matches. Instead of duct‑taping controls around exposed data, you keep the data protected by default and let policy travel with it. The effect is immediate: fewer plaintext paths, faster security approvals, and a cleaner way to enable AI without spraying secrets into prompts, embeddings, or vector stores. This lines up with enterprise reality: genAI usage is widespread and fragmented across hundreds to thousands of apps inside typical organizations, per Netskope’s Cloud & Threat Reports—which means protecting at the source beats chasing everything downstream.
What is Encryption-in-Use? Understanding Advanced Encryption Methods
Advanced encryption isn’t a buzzword. It is a practical triad: encryption at rest, encryption in transit, and encryption‑in‑use (also called searchable encryption). At rest and in transit protect stored media and the wire. Encryption‑in‑use extends protection to the exact moment data is most valuable and most at risk: while applications and AI agents are searching, matching, and processing it.
Paperclip SAFE represents the evolution of protecting an organization’s most critical and sensitive data. It keeps fields protected through policy‑aware queries and encrypted indexes so teams can perform permitted searches and joins without exposing raw values in application tiers, caches, logs, or data lakes. The outcome is better, faster access for the right users and systems without increasing complexity or sacrificing security.
Real-World Use Case: Secure Claims Processing with Encrypted Data
Consider a common workflow: a service agent must resolve a customer claim while an LLM assistant drafts the response. In the legacy pattern, identity data is decrypted inside the app, joined in a cache, logged for troubleshooting, and then partially redacted before the LLM sees it. Every step introduces exposure and slows approvals. Meanwhile, AI adoption isn’t waiting for you to perfect the guardrails—exactly the pattern highlighted by Cisco’s 2025 Data Privacy Benchmark.
With SAFE’s searchable encryption, the agent’s tool queries encrypted indexes and performs policy‑gated matches on protected fields. The assistant pulls only permitted context through the same policies, and its output never contained raw PII in memory. Your SLA holds. Your audit trail sings. Your plaintext footprint shrinks to almost nothing. This is how you square the circle of speed and safety, which is exactly where leadership focus is headed.
How Encryption-in-Use Addresses Top CISO Priorities
Speed without surrender. Boards want AI outcomes, not excuses. By keeping sensitive fields encrypted during processing, SAFE lets security teams approve more initiatives with less ceremony. That’s crucial when AI spending is scaling rapidly, as noted in Gartner’s 2025 outlook. Security must accelerate outcomes, not block them.
Complexity that finally simplifies. One policy model travels across databases, applications, and analytics stacks. Instead of bespoke tokenization and regex‑driven DLP for each team, you standardize sensitive‑data handling everywhere. That simplification matters as cyber risk becomes more interconnected, a theme emphasized in the WEF Outlook 2025.
Costs that make sense. Data breaches cost real money. Controls that eliminate plaintext in app tiers, caches, and logs reduce both likelihood and blast radius. IBM’s latest numbers show a global cost decline tied to faster containment, but the U.S. breach average remains painfully high. That makes reduction of exposure surfaces a fiduciary duty, not a nice‑to‑have, per IBM’s Cost of a Data Breach 2025.
AI data security that actually works. Encrypt at source. Govern at use. Field‑level policy decides what an agent can retrieve and how it’s revealed. SAFE’s encrypted indexes and field‑level policies ensure sensitive data is protected yet directly usable by permitted applications and AI agents — the protect‑and‑enable balance IDC calls out for leadership success. This minimizes the chance of accidental exposure across the sprawling constellation of AI tools already active in most enterprises—an adoption pattern documented by Netskope’s genAI telemetry.
Compliance as a by‑product. While only a small minority list compliance as their top pain point, you still need audit‑grade evidence without stalling delivery. Dual‑key governance, immutable logs, and crypto agility produce that evidence as part of everyday operations, which aligns with the spirit of the SEC’s incident disclosure regime.
Benefits by Executive Role: CISO, CIO, CEO, and Security Architects
For the CISO, SAFE eliminates the plaintext‑everywhere problem at design level. Policy follows the data rather than chasing it through ETL, caches, and analytics pipelines. This is data‑centric Zero Trust: the data never trusts the application layer to begin with. Your risk register shrinks and your security approvals speed up because you’re controlling exposure, not retrofitting controls around leaky systems.
For the CIO, SAFE is a software layer that respects your architecture. Integrate via standard APIs, keep schemas and SLAs intact, and stop writing custom masks and tokenization scripts for every team that touches sensitive data. Integration effort goes down while consistency and reliability go up. This is how you support business velocity without adding uncontrolled risk, a goal echoed by executive themes in the WEF Outlook.
For the CEO, this is about confidence. You can push forward on AI, analytics, and product innovation without gambling with reputation. When a regulator or major customer asks how you protect their information, you explain it simply: the data stays encrypted even while we use it. In a world where U.S. breach costs are still the highest globally, per IBM, that’s both brand protection and sound stewardship.
Network Security Architects get a posture that pairs neatly with Zero Trust architecture. TLS still protects the wire, segmentation still buys time, but with SAFE the payloads remain protected inside app boundaries and across east‑west traffic. Your SIEM gets cleaner signals from SAFE’s usage logs, and your egress rules become simpler because sensitive values aren’t escaping. The net effect is less brittle DLP at choke points and more durable, data‑centric control.
Technology Purchasing and Strategic Sourcing finally get a defensible, outcome‑based buy: cover the crown‑jewel fields, eliminate plaintext paths, and prove it with measurable SLOs. SAFE’s software‑first model avoids hardware lock‑in and lets you negotiate against outcomes rather than abstract headcounts or core licenses. That’s easier to justify to finance and aligns with the spend‑for‑outcomes trend documented by Gartner’s security spending outlook and Gartner’s AI spending forecast.
How Searchable Encryption Improves Data Access Without Adding Complexity
“Keep it encrypted” sounds like a slowdown. In practice, SAFE’s searchable encryption improves access by standardizing how sensitive data is used across teams. Here’s what changes:
- Search becomes safer and faster. Encrypted indexes avoid the decrypt‑match‑log‑redact cycle. Customer service, AML, and claims workflows can perform deterministic and policy‑bounded partial matches on protected fields. Answers arrive within existing latency budgets, and results are scoped to the role. This removes entire classes of caching and logging that used to be necessary evils.
- Joins don’t require secret staging. Common joins can be handled against SAFE‑compatible structures so you don’t materialize sensitive fields in staging tables or ETL scratchpads. Fewer moving parts equals fewer weak links. It also curbs the accidental exposure patterns that crop up in complex estates, the sort observed broadly in the DBIR.
- AI retrieval respects policy by construction. Retrieval‑augmented generation can pull only the permitted context, and outputs can be redacted, tokenized, or pseudonymized automatically. That helps contain the “shadow AI” problem seen in enterprise telemetry like Netskope’s reports.
- Operations simplify instead of sprawl. Policy‑as‑code, client‑managed or dual‑key custody, and crypto agility make control evolution manageable. This saves money and time, which shows up where it counts when incidents happen, as underscored by the correlation between faster containment and lower cost in IBM’s study.
90-Day Implementation Roadmap for Encryption-in-Use
Weeks 1–2: Pick two revenue‑adjacent workflows and draw the real map. Circle every place a sensitive field shows in plaintext: app memory, caches, logs, staging tables, exports. Meanwhile, define the handful of data domains that truly are the crown jewels. For risk framing, the WEF Outlook 2025 is a crisp way to explain why simplification is security.
Weeks 3–6: Apply SAFE to those domains with field‑level policies and dual‑key governance. Keep schemas and SLAs. Integrate the two workflows so queries and lookups run against encrypted indexes. Onboard SAFE audit events to your SIEM and write three detections: unusual index access, geographic anomalies, and off‑hours spikes on regulated fields. This is where you start compressing time‑to‑containment in the spirit of IBM’s findings.
Weeks 7–9: Wrap one AI workflow. If you’re doing retrieval‑augmented generation, make sure the context retrieval respects the same field‑level policies and that outputs are automatically redacted or pseudonymized where required. Run failure drills and validate performance. Publish the before‑and‑after: plaintext paths eliminated, SLOs met, approvals accelerated. This is the “security that accelerates” milestone your board wants.
Weeks 10–12: Expand coverage to a third workflow or domain, lock the production policy set in version control, and socialize the results to the board. Negotiate year‑one expansion with Procurement using outcomes: coverage achieved and time‑to‑approve improved. That’s how security spending aligns with the broader enterprise spend trajectories discussed in Gartner’s forecasts and the AI outlook.
Conclusion: The Future of Data Protection
If your strategy depends on data and AI, your controls should assume the data never needs to appear in plaintext to be useful. That is what Paperclip SAFE delivers: encryption‑in‑use as standard operating practice. Design once, use everywhere, move faster. The global trend lines point in the same direction: intertwined risk and escalating dependence on digital systems (WEF), record U.S. breach costs even as detection improves (IBM), and persistent credential‑centric attacks (Verizon DBIR). Encrypt at source. Govern at use. Keep moving.
Figure Credit
Source: IDC, WW C‑Suite Tech Survey 2025 (n=90); US53857425; Publication date Oct 9, 2025; Authors: Frank Dickson, Jennifer Glenn.