Attested Self-Modification for Autonomous Network Defense

A network defense that is configured once decays as its network and its adversary change, and the operator who must retune it by hand is always a step behind. We present an architecture for autonomous, co-evolutionary network defense in which the platform rewrites its own configuration, correlation logic, and protected surface, and an internal offense rewrites its own attacks to expose what the defense misses, with both selected continuously against a measured fitness signal so the platform hardens itself without human retuning. A defense that can rewrite itself can also disable itself, or be steered into doing so by an adversary who feeds it the signal that drives its adaptation, which is why self-modifying defenses are ordinarily kept under human control. The contribution of this paper is attested self-modification: every change the platform makes to itself is authenticated, before it is applied, by the dual-signature post-quantum protocol, pairing Ed448 with ML-DSA-87, that a companion paper introduced to authenticate the platform's management channel. A self-modification therefore carries the same provenance and the same resistance to forgery as a device identity, and an unauthorized or poisoned mutation is rejected by machinery the platform already runs. This generalizes the cryptographic-agility mechanism of the companion protocol, which authenticated the hot-swap of a single component, the cipher suite, to the mutation of the whole defense. The architecture further specifies quorum signing, a live second factor, and attack-triggered key rotation as hardening of the root of trust against key compromise, beyond the single dual signature the current implementation uses.