Autonomous telecom networks change the shape of the security problem. Agentic workloads move across models, tools, data stores, agents and infrastructure, carrying sensitive network intelligence with them. Paolo Campoli examines why protecting the infrastructure is no longer enough when the workload itself is designed to move.

Rome’s roads made the empire powerful by reducing the friction of distance, but the same infrastructure could be used by forces moving against it. Enterprise AI presents a modern version of that trade-off, concentrating in one environment the knowledge that was once scattered safely across separate systems.

Enterprises are making an increasingly unbalanced tradeoff with AI: proprietary knowledge is exposed to more infrastructure in exchange for capabilities that are becoming essential to compete. As AI concentrates valuable information into models and active computing environments, the potential cost of that tradeoff grows, demanding a fundamentally different security architecture.

Nadella calls it the Reverse Information Paradox: enterprises pay for AI twice, once in models and infrastructure, again in the proprietary knowledge they hand over each time they use it. But the root problem isn’t economic, it’s architectural: AI turns decades of security thinking on its head, forcing knowledge to become visible before it can be used.

For thirty years, perimeter security strictly limited data access in the enterprise. Agentic AI inverts this requirement entirely: agents need broad data visibility to function effectively, yet their distributed workflows span inference pipelines, retrieval systems, persistent memory, and inter-agent communication. Continuous data-level protection across all surfaces remains essential.

Researchers at KAIST demonstrated that AI model architectures can be reconstructed remotely through electromagnetic emissions from GPUs. ModelSpy exposes a growing gap in AI infrastructure security: confidential computing protects software boundaries, but not physical leakage. As sovereign AI and confidential AI accelerate, encrypted execution may become the next critical layer of AI defense.

The LiteLLM incident was a reminder that modern AI breaches often begin with the software supply chain. Once credentials are exposed, the real problem becomes control over access, environments, and downstream systems. That is where the blast radius grows, even when core assets remain protected.

Today, we anounced a significant milestone in DataKrypto’s mission to deliver the cryptographic foundation for Confidential AI: our completion of the Google Cloud ISV Startup Springboard Program and the availability of our flagship product, FHEnom for AI™, on the Google Cloud Marketplace.

On March 9, 2026, CodeWall.ai published a consequential AI security disclosure: “How We Hacked McKinsey’s AI Platform.” Codewall’s autonomous offensive agent — with no credentials, no insider knowledge, and no human guidance — achieved full read and write access to the production database behind Lilli, McKinsey’s internal AI platform used by more than 43,000 employees.

The market increasingly labels TEE-based architectures as “Confidential AI.” However, isolating an AI workload is not the same as securing the intelligence within it.

Trusted Execution Environments (TEEs)—such as Intel TDX, AMD SEV, and NVIDIA Confidential GPUs—protect the execution boundary, reduce infrastructure exposure, and limit host-level access to cloud workloads.

Security leaders are moving quickly to secure AI systems — often faster than the threat landscape affecting those systems is fully understood. Encryption is extended from existing data programs. Access controls are enforced. Monitoring is adapted. Governance processes are implemented. Much of this work is happening incrementally, which makes sense. AI adoption is still new, and compliance frameworks are evolving. As a result, many architectural decisions are made with uncertainty.

As AI becomes foundational infrastructure, traditional security controls are no longer enough. The real vulnerability emerges during computation itself — when sensitive data is decrypted in memory to be processed. Encryption-in-use closes this gap, enabling AI systems to operate on encrypted data without ever exposing it, making privacy a mathematical guarantee rather than a policy promise.

Artificial intelligence is rapidly becoming a central repository of enterprise knowledge. As organizations deploy private AI models trained on proprietary data, a new category of risk is emerging. This risk is not driven by how users interact with AI, but by how AI systems themselves are accessed, copied, and attacked.

Today, we’re sharing our 2026 Cybersecurity Predictions, offering our insights into the trends that we believe will be front and center in the coming year, particularly around AI and data protection. There is a great deal at stake. The business world is on the cusp of making significant AI breakthroughs, but the technology has some serious security gaps that must first be addressed.

Enterprises increasingly rely on large-scale cloud and accelerator infrastructure to run AI workloads. While encryption at rest and in transit are well understood, one part of the AI lifecycle remains comparatively underprotected: the moment data and model parameters are processed inside the accelerator.

It’s crazy to think that the world’s most-visited museum, the Louvre, could be breached in broad daylight. But that’s precisely what happened on October 19, 2025, when thieves disguised as workers used a basket lift to access the museum’s Seine-facing facade, broke through glass, and in just seven minutes stole eight priceless jewels from the Galerie d’Apollon.

Ransomware’s greatest weapon was never encryption — it was leverage. Stealing data and threatening public exposure gave attackers their bargaining chip. Fully homomorphic encryption eliminates that leverage entirely: when data stays encrypted even during processing, exfiltrated information becomes mathematical noise. No secrets, no leverage, no incentive to attack.

Last week, DataKrypto’s Head of Cryptography Research, Carla Mascia, spoke at Financial Cryptography in Rome 2025 (FCiR25), organized by the De Cifris Association with the support of Banca d’Italia, presenting a practical cryptographic toolbox showing how privacy and financial data analysis can not only coexist — but empower one another.

Six centuries ago, Leon Battista Alberti argued that only ciphers could safeguard secrets, because people could not. DataKrypto Founder and CTO Luigi Caramico draws a direct line from Alberti’s cipher wheel to modern AI security — where cryptography remains the only reliable way to engineer trust across untrusted systems.

Shadow AI is spreading fast, and banning it isn’t the answer. When employees can’t access approved tools that meet their needs, they turn to unsanctioned ones, creating invisible risk. DataKrypto Co-Founder and CEO Ravi Srivatsav explains how continuous encryption lets enterprises sanction more AI tools confidently, bringing shadow usage into the light.

Recently, Trail of Bits researchers, Kikimora Morozova and Suha Sabi Hussain, revealed in a blog post, a new class of attacks targeting multimodal AI systems—malicious instructions cleverly hidden inside images.

As AI takes center stage in the digital realm, every organization, from governments to national banks, is caught in a tug-of-war. On one hand, they need to use powerful AI tools. On the other hand, there are data sovereignty laws saying sensitive data can never leave the country.

Back in the 1980s, I was a hacker with a full head of hair — big, bold, and probably as unruly as the code I was exploiting. Those days shaped my relentless curiosity and drive to outsmart systems from the inside out.

As AI increasingly powers vital operations across industries, safeguarding AI workloads has become a top priority for CISOs facing a rapidly evolving threat landscape. Meanwhile, the looming quantum computing era threatens to break the very encryption methods protecting today’s data.

The Cybersecurity Infrastructure & Security Agency (CISA) recently published AI Data Security: Best Practices for Securing Data Used to Train & Operate AI Systems in collaboration with the National Security Agency’s Artificial Intelligence Security Center (AISC), the Federal Bureau of Investigation (FBI), the Australian Signals Directorate’s Australian Cyber Security Centre (ASD’s ACSC), New Zealand’s Government Communications Security Bureau’s National Cyber Security Centre (NCSC-NZ), and the United Kingdom’s National Cyber Security Centre (NCSC-UK).

As someone who has spent decades in the service provider industry, I’ve seen firsthand how technology can transform businesses and how quickly new opportunities can become new risks.

Widespread use of AI presents a double-edged sword: an ability to increase innovation and efficiency while expanding the risk of data exposure and misuse. That blade has created a dilemma that undermines the promise of innovation and efficiency, and makes companies hesitant to fully embrace AI due to concerns around protecting sensitive data and brand value.

Widespread use of AI presents a double-edged sword: an ability to increase innovation and efficiency while expanding the risk of data exposure and misuse. That blade has created a dilemma that undermines the promise of innovation and efficiency, and makes companies hesitant to fully embrace AI due to concerns around protecting sensitive data and brand value.

Massive data foibles are so common lately that they often seem like background noise. Yet, sometimes, an incident highlights vulnerabilities shared across countless organizations—especially in the rapidly expanding AI sector. One such case is the DeepSeek security lapse – although it might not have received the same level of attention had it not been for all the hype it received just a couple of days earlier.

Massive data foibles are so common lately that they often seem like background noise. Yet, sometimes, an incident highlights vulnerabilities shared across countless organizations—especially in the rapidly expanding AI sector. One such case is the DeepSeek security lapse – although it might not have received the same level of attention had it not been for all the hype it received just a couple of days earlier.

As crazy as it may seem, 2025 is upon us. As we look ahead to the coming year, we’re standing on the precipice of a new era in cybersecurity. It’s not just about defense anymore; it’s about innovation, about turning challenges into opportunities. We’re excited to share what’s top of mind for us as we head into a new year, based on conversations with our customers, technology leaders, and cybersecurity innovators.

As crazy as it may seem, 2025 is upon us. As we look ahead to the coming year, we’re standing on the precipice of a new era in cybersecurity. It’s not just about defense anymore; it’s about innovation, about turning challenges into opportunities. We’re excited to share what’s top of mind for us as we head into a new year, based on conversations with our customers, technology leaders, and cybersecurity innovators.

Like many aspects of cybersecurity, data security is a never-ending cat-and-mouse game between those trying to maintain continuous protection of data and those trying to gain unauthorized access for malicious purposes. Fortunately, tools are available that close data security gaps and make end-to-end protection across the data lifecycle a reality – eliminating blindspots and points of vulnerability for cyber defenders.

Data breaches are a relentless, escalating threat, costing companies millions and eroding trust worldwide. Traditional encryption methods are riddled with gaps and vulnerabilities (as we discussed in our previous blog), leaving organizations scrambling to understand and deploy alternative and complex data protection methods, such as trusted execution environments (TEEs) and confidential computing, which also have security gaps.

Gaps in data security exist – often because security leaders are unaware of the problem or assume there are no solutions. These gaps contribute to the escalating costs of data breaches, reaching an all-time high of nearly $5M per breach in 2024 (source: IBM, Cost of a Data Breach 2024).

Data is the organization’s crown jewel in the digital landscape, embodying the most valuable and sensitive information crucial to a company’s operations, success, and competitive edge. Safeguarding this data in its various forms—text, numbers, video, etc.—and across its lifecycle is essential to maintaining business integrity, protecting customers, and ensuring regulatory compliance.