deep-dive

What Independent Security Audits Find in Cloud Storage

A 2024 ETH Zurich study found theoretical weaknesses in pCloud's encryption. 66% of Azure accounts have risky defaults. Here's what auditors keep uncovering.

In 2024, researchers at ETH Zurich published a paper examining the encryption architectures of five major cloud storage services. For pCloud, they identified a class of theoretical attacks in which a compromised server could insert files into a user’s storage or silently swap the encryption keys used to protect content — without the user being able to detect it. pCloud called these attacks theoretical and said they require a compromised server, which implies a serious breach of pCloud’s own infrastructure. Researchers responded that “theoretical” is precisely the correct level at which to evaluate encryption: if a vulnerability exists in the design, the protection is weaker than claimed, regardless of whether it has been exploited.

Then, in February 2026, users reported a different issue: some pCloud users were seeing file names belonging to other accounts in their own storage. pCloud acknowledged the bug was real. The researchers said it was consistent with the key-management weaknesses they had identified.

This is not an isolated event in the world of cloud storage security research. Independent audits and academic studies of cloud storage security consistently find the same categories of problems: architecture decisions that undermine the encryption they describe, configuration defaults that expose data unnecessarily, and gaps between what marketing materials claim and what the technical implementation actually provides.

Understanding what auditors look for — and what they keep finding — is useful for anyone who has weighed privacy when choosing where to store personal files.


What Independent Audits Actually Examine

When security researchers audit cloud storage, they are generally not trying to hack a service in a legal sense. They are examining the architecture and implementation decisions to determine whether the security properties the service claims to provide — encryption, access control, data isolation — actually hold up under scrutiny.

The main categories of examination:

Encryption architecture. Does the service encrypt files at rest? Who holds the encryption keys? If the service holds the keys (server-side encryption), it can decrypt your files. If you hold the keys and the service never sees them (client-side or zero-knowledge encryption), the service cannot read your content even if compelled. Most consumer cloud storage services use server-side encryption — the service holds the keys.

Key management. Even with encryption present, how keys are managed matters enormously. A service might encrypt files at rest but store the encryption keys in a way that is technically accessible to a compromised server, a compromised employee, or a law enforcement request that compels key disclosure. The ETH Zurich paper’s finding about pCloud was fundamentally a key management finding: the design allowed key substitution under adversarial conditions.

Encryption in transit. Is data encrypted between your device and the service’s servers? Most reputable services have used TLS for transport encryption for years. The relevant question now is which TLS version — TLS 1.2 has known weaknesses; TLS 1.3 is current — and whether certificate validation is implemented correctly.

Access control and tenant isolation. If the service stores multiple users’ data on shared infrastructure, can one user’s data reach another user? The pCloud file-name visibility bug was a tenant isolation failure: under some conditions, data that should have been visible only to one account was visible to another.

Authentication and session security. How does the service authenticate users? Are authentication tokens issued with appropriate scope and expiration? Can a compromised token give broader access than the user intended?


What Audits Keep Finding

The research literature on cloud storage security, and the breach reports that continue to emerge, point to several recurring problems.

Misconfigured default settings. A 2026 assessment of Microsoft Azure deployments found that 66% of Azure accounts had storage account access keys enabled — a legacy authentication mechanism that provides broad access without modern controls — and 61% allowed public network access to storage accounts. These are not the result of malicious configuration; they are defaults that were never reviewed and changed.

Default settings represent the security posture of users who did not configure anything. Cloud services set these defaults, and the defaults determine how secure most accounts actually are. An encryption feature that exists but is off by default protects only the users who found and enabled it.

The gap between “encrypted” and “protected.” Encryption is a tool, and its protection depends on implementation details that marketing claims do not always describe accurately. “Your files are encrypted” and “your files are encrypted in a way that prevents us from reading them” are meaningfully different claims. The first describes a technical mechanism. The second makes a claim about access that depends on how the keys are managed.

Research on cloud storage services has repeatedly found this gap: files are encrypted, but the keys are held by the service, accessible via internal systems, and disclosable in response to legal process. This is not fraud — it is often disclosed in technical documentation — but it is not what most users mean when they hear “encrypted.”

Third-party component vulnerabilities. Many cloud services use third-party software components for specific functions — file preview, media processing, format conversion. These components introduce additional attack surface. A vulnerability in a PDF preview library can give an attacker access to the content of any PDF a user views, regardless of how well the primary storage layer is secured.

Insufficient logging and detection. Several cloud incidents have unfolded over extended periods — months — before detection, because the services did not have sufficient logging or anomaly detection to identify unusual access patterns. The California student data breach in 2026 involved unauthorized access over a four-month window (May through September) before it was identified.


Server-Side Versus Client-Side Encryption: The Practical Difference

Understanding this distinction is essential for evaluating cloud storage security claims accurately.

Server-side encryption (also called encryption at rest) means that files are stored on the service’s servers in encrypted form. The encryption keys are also stored or accessible by the service. The file is decrypted when you access it, and the service handles the decryption. This protects against an attacker who gains physical access to storage media — they get encrypted data, not readable files. It does not protect against legal requests to the service, compromised service infrastructure, or malicious insiders who have access to the key management system.

This is the model most consumer cloud storage services use, including major providers. It is meaningfully better than no encryption. It is not the same as the service being unable to read your files.

Client-side encryption (also called zero-knowledge or end-to-end encryption) means that files are encrypted on your device before they are uploaded. The service stores encrypted data that it cannot decrypt — it does not hold the keys. This protects against legal requests to the service (they have nothing to hand over), compromised service infrastructure, and malicious insiders. It does not protect against malware on your own device or the loss of your own encryption keys.

A small number of cloud storage services offer client-side encryption as their primary model. This is a stronger privacy guarantee, but it comes with trade-offs: if you lose your encryption key, your files are inaccessible permanently. Features that require server-side processing — AI-powered search, automatic album organization, document preview — generally cannot work with client-side encrypted data because the server cannot see the content to process it.

Most consumer cloud storage services, including those marketed as privacy-focused, use server-side encryption. This includes pCloud in its standard configuration, Google Drive, iCloud, Dropbox, and OneDrive. Some offer client-side encryption as an optional, additional feature — but not as the default.


The Audit Gap for Consumer Products

Enterprise cloud storage receives regular independent audits because enterprise buyers require them. SOC 2 Type II reports, ISO 27001 certifications, and penetration testing reports are standard in B2B cloud procurement. These do not guarantee perfect security — a service can be SOC 2 certified and still have incidents — but they represent independent verification that specific security controls are in place.

Consumer cloud storage products receive far less independent scrutiny. The ETH Zurich paper on cloud storage encryption is notable partly because academic security research targeting consumer cloud services is rare. Most assessment of consumer cloud storage security comes from journalists, independent researchers, and, unfortunately, breach disclosures after the fact.

This creates an information asymmetry: enterprise buyers can inspect audit reports; individual users generally cannot. What you can do is look for services that:

  • Publish their technical security documentation, not just marketing copy
  • Have disclosed past incidents transparently and promptly
  • Have been the subject of independent security research (even critical research is more useful than no research)
  • Make clear, technically specific claims about encryption architecture rather than general assurances

Reading Security Claims Accurately

When evaluating a cloud storage service’s security claims, a few questions cut through the marketing language:

Who holds the encryption keys? If the answer is not specified or involves language like “your files are secured with military-grade encryption,” the answer is almost certainly the service itself. Look for explicit statements about key custody.

Is there an audit? A published SOC 2 Type II report, a named independent security auditor, or published security research findings indicate more scrutiny than marketing claims alone.

What encryption protocol and version? Specific answers — “TLS 1.3 in transit, AES-256 at rest” — are better than vague ones. AES-256 is a standard at-rest encryption level for reputable services. TLS 1.3 is the current standard for transport encryption.

Has the service had incidents, and how were they handled? Past incidents handled with transparency and timely disclosure are a more useful signal than no reported incidents, which may simply mean inadequate security monitoring.


The Honest Position

Most reputable cloud storage services offer meaningful protection against some threats — physical media theft, casual unauthorized access — while not offering protection against others, including legal process, compromised infrastructure, and serious insider threats. This is not a failure; it is an honest description of what server-side encryption provides.

The problems that audits keep finding — misconfigured defaults, key management weaknesses, insufficient logging, tenant isolation failures — are solvable problems. Some services have solved them better than others. The gap between “encrypted storage” as a marketing claim and encryption as a meaningful privacy protection is real, but it narrows significantly for services that have invested in their security architecture and been transparent about what they offer.

daftei uses TLS 1.3 in transit and AES-256 encryption at rest — server-side encryption with industry-standard parameters. This is the honest description of what is provided: protection against unauthorized access at the storage layer, without the claim of zero-knowledge encryption that would make server-side AI features impossible. Files are never analyzed for AI training, never sold, and never used for advertising. Account deletion triggers permanent, irreversible erasure after a 30-day grace window.

Understanding what those specific claims mean — and what they do not mean — is more useful than any marketing summary.

Your memories deserve better than an ad platform.

Try daftei free →
← All posts