Two companies, one focused on encryption and the other on AI trust, have joined forces to deliver what they call the world’s first fully encrypted guardrails for AI. In an age of accelerating model adoption, regulatory pressure and constant AI drift (not to mention AI Shiny Object Syndrome), their collaboration marks a turning point for how enterprises approach and operationalize AI security.
The announcement, led by encryption innovator DataKrypto and AI governance platform Tumeryk, introduces Encrypted Guardrails for Operational AI Security, a layered solution designed to secure every step of the AI pipeline, from retrieval-augmented generation (RAG) and model weight protection to prompt inspection and real-time compliance alignment. “DataKrypto is proud to partner with Tumeryk to set a new standard for secure AI adoption,” said Ravi Srivatsav, co-founder and CEO of DataKrypto. “This collaboration empowers organizations to deploy scalable AI workloads — even on their most sensitive data — without ever compromising privacy or performance.”
On paper, it’s a strategic integration. In practice, it’s a philosophical handshake between two fast-rising companies that share a bold thesis. If AI isn’t secured at the operational level, deep in the infrastructure where data flows and prompts are processed, it’s only a matter of time before filters fail. “Encrypted Guardrails close the final mile of AI security,” said Rohit Valia, founder and CEO of Tumeryk. “Organizations can now innovate with GenAI, confident that every token, from retrieval to response, is both policy-aligned and cryptographically protected.”
A Partnership Decades in the Making
The spark for the partnership wasn’t a sales call or a VC pitch; it was a personal connection. Paolo Campoli, a longtime executive at Cisco and founder of Campoli Consulting, had known Luigi Caramico, co-founder, CTO and chairman of DataKrypto, for nearly 30 years, from the early days of interactive television and broadband fiber deployment across Italy. “Luigi and I built our careers on trust, on networks that worked when others failed,” Campoli said. “When I saw what Tumeryk was doing with their AI policy engine, I knew it would pair perfectly with DataKrypto’s always-on encryption. It wasn’t a matter of if. It was how fast we could get the right people in the room.”
Campoli introduced the two teams, helped translate their shared goals, and watched as their platforms “locked into place.” His instincts were right. The integration offers protection and observability at every level of the AI stack, precisely what enterprise customers and regulators are demanding.
Today, Campoli serves as a strategic advisor to both firms, helping bring the solution to telcos, governments and managed services providers (MSPs) across Europe, the Middle East and Africa. “This isn’t just a technical innovation, it’s a regulatory unlock,” he said. “European MSPs can now offer secure, validated, notarized AI workloads. That’s not a product, it’s a lifeline.”
An Innovative Security Redesign For Traditionally Weak Guardrails?
Enterprises racing to adopt large language models (LLMs) are waking up to a hard truth: The very AI systems that offer breakthrough productivity gains can also introduce unprecedented risk, especially if deployed without deep technical safeguards and policy-aware controls.
Historically, AI guardrails have focused on outputs, flagging toxic language, harmful instructions, or non-compliant responses after the model generates them. But that leaves upstream vulnerabilities untouched. Data payloads, vector embeddings and even the way a prompt is structured, these remain wide open to prompt injection, model inversion and insider misuse.
The joint offering encrypts everything from RAG queries and embeddings to prompts and policy payloads. This goes beyond protecting data at rest or in transit, using trusted execution environments (TEEs), the system supports encrypted computation in use. Prompts are inspected in real time inside secure enclaves before they reach the model, ensuring user intent aligns with policy at every step.
“We don’t just hand you a badge and let you wander the building,” said Caramico. “Tumeryk ensures you behave once inside. It’s like a courtroom that checks your ID at the door and then monitors your actions once inside, policy by policy, token by token to ensure security and following of established rules.” For Caramico, this partnership solves a critical gap. “Our job has always been to keep unauthorized users out,” he said. “But traditional systems stop there. They don’t ask what an authorized user is actually doing in the environment. That’s where Tumeryk changes the game.”
The DataKrypto-Tumeryk solution encrypts the entire chain of AI operations, leveraging DataKrypto’s FHEnom™ encryption engine and Tumeryk’s policy-aware prompt inspection. Prompts are evaluated in real time inside TEEs, ensuring encrypted context windows never leave their trusted domain. At no point is the data unwrapped and exposed to the outside world, even during active processing.
The result is an environment where even the guardrails themselves are tamper-proof, encrypted, version-controlled and anchored to runtime policies that can’t be manipulated mid-process. “You can’t secure the future by fixing yesterday,” Caramico said. “If the foundation is cracked, the whole building’s at risk. We’re not just reactively plugging leaks, we’re proactively building a watertight system from the inside out.”
And because Tumeryk’s telemetry maps events directly to industry frameworks like NIST AI RMF, ISO 42001, PCI DSS and HIPAA, security teams can prove policy enforcement with real-time dashboards, turning AI risk from a black box into a reportable metric.
“We’re finally at the point where AI guardrails aren’t just ethical theater,” Campoli said. “They’re enforceable controls with traceable compliance.”
The Sovereign Cloud Use Case
The demand for operational AI security isn’t abstract; it’s geopolitical. In Europe, the forthcoming EU AI Act is forcing organizations to demonstrate not just compliance, but provable control. DataKrypto’s infrastructure, designed to run independently of U.S. hyperscalers, offers telcos and public sector organizations a way to build sovereign AI systems without compromising on speed or scalability.
Caramico says the early interest from EMEA-based MSPs, banks and infrastructure operators has been overwhelming. “Sovereignty isn’t just a buzzword, it’s a business requirement,” he said. “This partnership gives organizations the confidence to offer AI as a service, knowing every token is protected and every policy is enforced.”
Campoli echoed the sentiment. “The feedback we’re hearing is crystal clear: ‘We want to use AI, but only if we can prove the data stayed under our control.’ This gives them that assurance.”
Trust by Design, NOT as an Afterthought
What makes this integration stand out isn’t just its technical depth; it’s the philosophy behind it. Both Caramico and Tumeryk CEO Rohit Valia have expressed frustration with the security industry’s reactive posture toward AI.
Too many companies, they say, are waiting for a breach, a regulation, or a scandal to act. The DataKrypto-Tumeryk model is proactive by design — baking in encryption, observability and red-team resistant controls from the start. “We’re not here to bolt a seatbelt onto a race car after it’s crashed,” Caramico said. “We’re here to build the roll cage into the frame and protections for the drivers.”
For Campoli, it’s about trust, earned, maintained and never assumed. “If we want AI to be trusted, it must earn that trust in every interaction. You can’t fake it, you can’t win it with a marketing hammer, and you can’t audit it into existence after the fact. What DataKrypto and Tumeryk have built makes trust a technical property, not just a marketing slogan.”
Top 5 Capabilities from the DataKrypto/Tumeryk Encrypted Guardrails Partnership
- Always-On Encryption for Every AI Layer
Powered by DataKrypto
Applies Fully Homomorphic Encryption (FHE) across raw data, embeddings, model weights, prompts and responses — ensuring all data remains encrypted, even during runtime. - Trusted Execution Environments (TEEs) with Real-Time Prompt Security
Joint Integration
Encrypted prompts are evaluated inside secure enclaves. Tumeryk’s adaptive guardrails block toxic or non-compliant requests without breaking encryption boundaries. - AI Trust Score™ with Live Compliance Mapping
Powered by Tumeryk
All telemetry maps to frameworks like NIST, ISO 42001, and SOC 2. Compliance and risk insights are available in real time, with auditable enforcement logs. - Tamper-Proof Guardrails and Policy Enforcement
Joint Innovation
Policies are encrypted, version-locked and runtime-anchored, preventing manipulation even from insiders or privileged users. - Plug-and-Play Support for GenAI Workloads and AI Agents
Delivered by Both Teams
Supports both large and small language models (LLMs and SLMs) with native integration. Future-ready for agent swarms and autonomous orchestration.
Available Now with Eyes on What’s Next
The solution is now available as a managed SaaS or self-hosted deployment. Pricing is available upon request, and live demos can be scheduled via either company’s website. The roadmap includes deeper support for SLMs, edge deployments and decentralized agent systems, ensuring that the integration remains viable even as AI evolves.
“We’re not solving yesterday’s problems,” Caramico said. “We’re building and securing the rails that tomorrow’s AI will travel on.”
