Answers are grounded in this paper and cite their sources.
Information-theoretic deletion of one-time-pad ciphertext reduces to the destruction of the pad — provided the pad was information-theoretically secret to begin with, which means it came from a true random source whose raw output has been independently assessed, and no seed or generator state that could regenerate it exists. This paper examines what that reduction actually costs on the devices this class of system runs on, and reports a negative result: on flash storage managed by a translation layer, an application cannot guarantee that a specific record's bytes are destroyed. The translation layer performs wear levelling, so a logical overwrite is a write to a fresh physical page plus a remapping, leaving the prior contents resident and unaddressable rather than absent. Every commonly proposed mitigation — repeated overwriting, discard hints, filling free space, device-wide secure erase, platform storage encryption — fails to close this for per-record deletion, and each fails for a different and instructive reason. We give a construction that moves the dependency onto a much smaller object: pad material is never written in the clear, but sealed at rest under a random key of its own per slice. Destroying that key leaves any surviving remnant protected computationally, so the deletion guarantee drops from information-theoretic to computational, and it holds only as far as the key's own destruction holds. We state the resulting guarantee precisely, including the adversary against whom it gains nothing without hardware support; describe the implementation rules that make the property real rather than nominal, including one that a natural design — and our own published companion paper — violates; report what a key-attestation measurement on two consumer handsets does and does not show; describe the discipline by which an implementation is kept from strengthening its own wording over time; and identify the consequences for a family of related claims that are commonly stated too strongly, including our own. Keywords: secure deletion, cryptographic erasure, crypto-shredding, flash translation layer, wear levelling, flash forensics, one-time pad, honest claims. ---
No takes yet. Share an insight, caveat, or question.
AriaLabs Systems LLC (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: