Persistent Computational Residue from Autonomous AI Agents

Autonomous agents can alter environments that outlive their initiating processes. Files, databases, caches, queues, credentials, memory stores, learned procedures, and physical device states can subsequently influence other computations. Brown’s computational-residue hypothesis asks whether these changes can accumulate, interfere with successor agents, propagate between systems, or produce longer-term degradation. The scientific difficulty is distinguishing known retained state from an additional phenomenon that existing mechanisms do not explain. This investigation reviews historical work on software aging, stigmergy, distributed state and termination, together with contemporary research on persistent agent memory, contamination, and recovery. A version-specific register contains 56 scientific, technical, and author records; a focused analysis examines 12 patent publications. Accessible full texts were reviewed at the sections specified in the register; several historical sources remain abstract-only. The synthesis supports persistence and downstream influence through explicit carriers, but does not establish a new agent-specific substrate or a general process of Internet-wide degradation. Particularly close earlier proposals already describe fresh-agent succession and selective environmental resets. Eight hypotheses separate persistence, accumulation, additional effects, successor interference, propagation, recovery, generalization, and long-term harm. Conditional state-transition and accumulation models clarify what can be identified, while a complete-state restoration proposition explains why a historical autonomy label cannot independently affect future outcomes when all causally relevant state and transition laws are identical. Fourteen proposed experiments combine matched replay, verified cessation, selective resets, controlled transfer, and independent replication. Negative results require sensitivity and equivalence bounds. The resulting contribution is an evidence-qualified framework for investigating environmental carryover, rather than a claim that unexplained computational residue has been discovered.