Data security in the cloud now faces the challenge of protecting data in use, at the exact moment it is processed by applications, analytics and AI models. That is because the most critical vulnerability sits precisely at runtime: when a piece of data is decrypted to generate value, it can in theory be exposed.
Confidential Computing was created to close this gap, isolating sensitive information in protected hardware environments even during processing. A key technology for cloud security, AI governance and regulatory compliance in an increasingly regulated context. Let’s take a closer look.
What is Confidential Computing?
Confidential Computing is an approach to security designed to protect data at the most critical moment of its life cycle, that is when it is being processed, not stopping at the encryption of data “at rest” or “in transit”, but stepping in at runtime, in other words when applications, AI algorithms or analytics systems access the information.
In the traditional model, when an application has to process a piece of data, that data is decrypted in the system’s memory. Even if only for a few milliseconds, in that window it can in theory be exposed to memory dumps, privilege escalation, hypervisor compromise or insider threats. It is a structural risk of conventional software architecture.
Confidential Computing removes this vulnerability by moving the security perimeter directly into the hardware. It does so through Trusted Execution Environments, isolated enclaves inside the CPU where code and data are executed in a protected way. Inside these secure areas, the information stays invisible and inaccessible not only to other processes, but also to the operating system, the hypervisor and even the cloud provider.
The goal of Confidential Computing is therefore to ensure that data stays encrypted and protected even while it is in use, drastically reducing the attack surface in cloud and multi-tenant environments.
In 2019 the Confidential Computing Consortium was set up under the aegis of the Linux Foundation, bringing together technology companies, cloud providers and hardware manufacturers with the aim of defining common standards and promoting interoperable open source tools. Its members include organisations such as IBM, Microsoft and Google.
With Confidential Computing, then, trust is no longer based exclusively on software and access policies, but is anchored to verifiable cryptographic and hardware isolation mechanisms.
Isolation, encryption and attestation: how Confidential Computing works
Confidential Computing combines three fundamental elements: hardware isolation, advanced encryption and cryptographic verification of integrity.
When an application has to process sensitive data, the data is loaded into a protected enclave, namely the TEE. Isolation is guaranteed directly by the processor: not even the operating system, the hypervisor or an administrator with root privileges can access the contents in memory. The cryptographic keys are managed at hardware level and remain accessible exclusively to authorised code. If the code is altered or if an unauthorised process tries to interact with the enclave, access is blocked automatically.
A central role is played by remote attestation: the system can produce a cryptographic proof certifying that the code is authentic and has not been tampered with. A mechanism that lets partners, customers or external systems verify that a given workload is running in a genuinely secure environment.
Today this aspect is crucial above all in multi-cloud contexts and in digital supply chains, where trust cannot rest on contracts or policies alone, but has to be technically demonstrable.
Why Confidential Computing matters today
The explosion of AI has radically changed the security landscape, because AI models are trained and used on increasingly sensitive datasets (financial, healthcare, industrial, government and so on). Without protection in use, this data has to be decrypted during processing, increasing the risk of exposure.
Confidential Computing, by contrast, makes it possible to run inference, analytics and training while keeping the isolation in place even at runtime. A system that is particularly relevant for:
- Generative AI in the enterprise
- Data collaboration between companies
- Federated training
- Protection of proprietary models
If we consider today’s European legislative context, ever more attentive to AI governance (which will become fully applicable this year), Confidential Computing will become a strategic tool for innovating without violating the principles of minimisation, security and accountability.
Indeed, increasingly robust technical and organisational measures for data protection are already required, since the GDPR, the Data Act and the EU AI Act impose strict controls on access to, processing of and transfer of information. Confidential Computing aligns naturally with these principles because it:
- Reduces the attack surface
- Prevents unauthorised privileged access
- Provides verifiable attestation mechanisms
- Supports digital sovereignty models
Hardware isolation makes it possible to demonstrate that data has been processed in secure environments even on shared infrastructure, a particularly important point for organisations operating in hybrid cloud or multi-tenant environments. In addition, many confidential computing platforms include logging and audit features that make it easier to demonstrate compliance during an inspection.
Despite the advantages, Confidential Computing is not without complexity, because integration may require changes to the applications, especially for legacy workloads. Performance, although it has improved in recent years, can suffer from hardware isolation in some specific cases. In addition, key management and orchestration at scale require advanced skills in security and cloud architecture. For this reason, effective adoption has to be part of a security-by-design strategy, not a mere additional layer.
Concrete applications in different sectors
Confidential Computing is already built into many enterprise architectures. In the financial sector, for example, it makes it possible to process transactions and risk models in public cloud environments while keeping customer data protected. In healthcare, it makes it possible to analyse medical records or genetic data for research and personalised medicine without compromising patient privacy. In the industrial and IoT world it protects the data generated by machinery and sensors, especially in edge computing scenarios where the infrastructure may be physically exposed.
One strategic area is the protection of intellectual property: algorithms, AI models and business logic can be executed in shared environments without the risk of copying or reverse engineering.
Adopting Confidential Computing brings advantages that go beyond cybersecurity, strengthening trust in cloud services and so making it possible to choose providers on the basis of technical and business criteria, not just perceived risk. It also enables new models of collaboration between companies, which can share computational results without sharing raw data or proprietary algorithms. From an economic point of view, better protection of sensitive workloads reduces exposure to regulatory penalties and reputational damage.
Strategy: adopting Confidential Computing
Adopting Confidential Computing inevitably leads a company to rethink its security architecture in strategic terms. Effective integration starts from identifying the workloads that are truly critical and continues with a careful assessment of the hardware and cloud platforms compatible with its operational needs.
Not all TEEs offer the same characteristics in terms of isolation, performance, attestation and key management. This is why adoption has to be guided by an architectural vision: integration with hybrid and multi-cloud environments, compatibility with DevSecOps pipelines, orchestration at scale and continuous monitoring.
The adoption of Confidential Computing should start from a clear analysis of the risks and of the strategic assets. Once the priority cases have been identified, it is essential to verify the compatibility of the infrastructure, such as hardware support for TEEs, availability of confidential cloud solutions, integration with key management systems and DevSecOps pipelines.
The next step is to launch a targeted PoC, on a single high-impact service or application, to assess performance, operational management and the real benefits in terms of risk reduction.
Only after this phase can adoption be extended progressively, integrating Confidential Computing into security governance and into continuous monitoring processes.
Confidential Computing does not replace other security measures such as identity management, traditional encryption and network segmentation, but strengthens them during processing.
The real turning point, however, is cultural, because moving trust from software to hardware means enabling a trust-by-design model, in which security is cryptographically verifiable and not based exclusively on policies or administrative controls. This allows companies to innovate with greater freedom, with a foundation on which to build a digital infrastructure capable of generating value without compromising trust, control and responsibility.
Do you want to turn the security of your data into a competitive advantage? dotenv does not just develop custom software, it listens to the needs of your business and designs complete technology solutions that bring together security, scalability and innovation in the cloud.
We support companies at every stage as they face the challenges of digitalisation, protecting critical data and strategic workloads.