Digital Trust under Autonomous AI and Quantum Risk: Signed Memory, Cryptographic Unpredictability, and Harvest-Now-Decrypt-Later

Governments are adopting artificial intelligence at the same moment that the cryptographic assumptions under their long-lived secrets are being placed on a clock. This paper treats four developments as a single problem of digital trust rather than as separate technical topics. First, AI systems are moving from content generation to agentic action, and are already exposed to impersonation, poisoning and unauthorised use. Second, persistent memory across sessions creates a new class of institutional record, whose integrity cannot be reduced to a signature. Third, cryptographic security depends on unpredictable keys, not on non-determinism, and better randomness does not answer a quantum attack on public-key mathematics. Fourth, harvest-now-decrypt-later means that the threat to data which must remain confidential for decades exists before a cryptographically relevant quantum computer does. Drawing on published resource estimates, the NIST post-quantum standards, and the positions of major security agencies on quantum key distribution, the paper argues that preparation should begin with an inventory of vulnerable cryptography, hybrid migration of high-value key exchange, and an explicit regime for the ownership and correction of AI memory. The implications for diplomatic communications and for states with constrained compute and language resources are developed throughout. KSCCN discussion paper prepared for the SQLL session on AI and quantum computing, Baridhara DOHS, 3 October 2026. Dr. Syed Muntasir Mamun, ORCID 0000-0001-6845-2853. The views expressed are those of the author.