Answers are grounded in this paper and cite their sources.
On the most widely deployed mobile platform, text entry runs through a replaceable, user-selected input method that receives every keystroke — including keystrokes typed into fields marked secret. Every passphrase-based scheme begins with the sentence "the user enters the passphrase," and on such a platform that sentence conceals a third party inside the trust boundary: a program the application did not write and cannot inspect, which by architectural necessity observes every keystroke before the application receives any of them. We characterise the resulting input-path leakage for high-value secrets and show that, on such platforms, the usual input-method mitigations are requests rather than enforced controls — addressed to the very component whose behaviour is in question. (At least one other major platform enforces part of this itself: it substitutes its own keyboard in secret-marked fields and lets an application refuse third-party keyboards, which moves trust to the platform vendor's keyboard rather than removing it.) We give a construction that removes the input method from the trust boundary entirely: an application-drawn fixed character grid, live prefix filtering against a fixed dictionary, candidate completion with no predictive model and no state outliving the session, and a passphrase generated by the application rather than chosen by the user. We report the entropy consequence of generation versus choice, including what free rerolling costs; discuss key-derivation parameter selection against the low-cost hardware this class of software exists to serve; and delimit what the construction does not buy. The input-method threat is well documented in the mobile-security literature, and defences exist, generally requiring hardware support or platform cooperation; dedicated signing devices already pair generated word secrets with on-device entry. The gap we address is narrower: the cryptographic literature on pad- and passphrase-based schemes specifies at length what happens to a secret once an application holds it — sealing, storage, erasure, attestation — while conditioning all of it on a first step that the mobile-security literature has been documenting as broken for a decade. Keywords: input method editor, keylogging, trusted input, passphrase entropy, key derivation, mobile security, threat modelling. ---
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: