A new class of hardware attack, termed DDRop, has exposed a fundamental weakness in the confidential computing defences that both Intel and AMD have built into their latest processors. The attack works by silently dropping memory writes, forcing the processor to continue reading stale encrypted data as though it were current. For operators deploying confidential computing across their infrastructure, this disclosure highlights the gap between architectural assumptions and physical reality.
How DDRop Undermines Confidential Computing
Confidential computing platforms like Intel's TDX (Trust Domain Extensions) and AMD's SEV-SNP (Secure Encrypted Virtualisation–Secure Nested Paging) are designed to encrypt sensitive workloads in memory, protecting them even from privileged software and hypervisor access. The threat model assumes that the memory subsystem itself is trustworthy—that data written to RAM actually reaches its destination, and that reads return the data that was written.
DDRop exploits this assumption by manipulating memory writes at the physical level. When an attacker with existing software access and brief physical access to a server instals a small circuit in the memory path, they can selectively suppress write operations. The processor's cache coherency logic then serves stale data from its caches, allowing an attacker to observe or influence what the confidential workload reads without triggering memory protection exceptions. The encrypted data remains encrypted, but it's data from a previous state—effectively a replay attack at the hardware layer.
The Attack Surface and Threat Model
The attack is not a remote exploit. It requires an adversary who can both control the host software and physically access the server hardware—a plausible scenario in colocated facilities, cloud provider supply chains, or insider threat scenarios. For organisations running sensitive workloads in hosted environments, this expands the threat surface beyond software-level compromise.
What makes DDRop particularly concerning is its silent nature. Unlike a crash or exception, suppressed writes leave no obvious trace. A workload may continue executing, reading corrupted or replayed state, without alerting monitoring systems. This makes detection difficult unless operators implement memory access telemetry that is itself resilient to hardware-layer manipulation.
Implications for Hosted Confidential Computing
For infrastructure providers offering confidential computing instances, DDRop underscores the importance of supply-chain security and tamper detection. Hosting operators must ensure physical security controls are sufficiently rigorous that circuit-level modifications are detectable. This includes:
- Sealed, monitored DIMM slots and tamper-evident hardware enclosures
- Regular inventory audits of memory modules and interconnect hardware
- Compartmentalisation so that a single server compromise does not expose multiple customer workloads
Additionally, operators should assume that confidential computing is a layered defence. Memory encryption protects against data centre staff reading unencrypted RAM, but hardware attacks like DDRop remind us that cryptographic protection alone cannot account for physical-layer manipulation. Operators deploying confidential computing should understand these limitations and design their threat models accordingly—perhaps restricting confidential instances to customers with commensurate risk profiles, or requiring attestation mechanisms that verify the absence of known hardware modifications.
The Broader Picture
DDRop is one of a series of hardware vulnerabilities that have chipped away at confidential computing's security promises over the past few years. Prior attacks have exploited side channels, cache behaviour, and I/O interfaces. Each disclosure narrows the assumption space on which these defences rely.
For operators, the takeaway is clear: confidential computing is a useful tool for protecting sensitive workloads, but it is not a substitute for other security controls. Proper compartmentalisation, access controls, and monitoring remain essential. Hardware attacks like DDRop are a reminder that infrastructure security is not solely a software problem; it's a systems problem that spans physical security, supply-chain integrity, and cryptographic guarantees.
