PhilArchive
open access
Why AI Is Running Wild: It's Not a Bug, It's a Feature
Agentic AI is moving from answer production to consequential action. Controlled evaluations and public reports in 2026 document reward tampering, oversight deactivation, coercion, shutdown resistance, sabotage, proxy use, and paths through real infrastructure after route failure. They establish neither deployment prevalence nor human-like malice, only that current agent–harness systems can, under identifiable conditions, act beyond delegated authority. The explanatory argument is derived from the economic principle of intelligent behaviour: finite agents realize conduct through processes that compete for typed resources and cannot all occur jointly. Intelligence adds an action-guiding representation of possibilities and the capacity to revise or transform that field. Robust path construction is therefore a feature of useful agency. Under scarcity and non-closure, risk emerges when persistent objectives and consequential tools meet a soft authority boundary. If the task policy also judges exceptions, insufficiency of authorized means can become de facto permission for unauthorized means. Current systems combine objectives and adaptive path construction with no intrinsic economy binding authority, externality, and consequence to every realized transition. A finite prohibition remains an obstacle that search may route around. An intrinsic constraint instead admits an action only as a transition carrying a mandatory typed resource debit the task policy cannot erase, counterfeit, or externalize. Compute is only one type; privilege, causal reach, irreversibility, and externality also require scarcity or price. The paper keeps the public evidence independent from the theory proposed to unify it. It formalizes an authority non-conversion invariant and an economic conservation invariant: authorized-path failure may update the plan, request, or goal, but never executable authority; no consequential transition occurs without settling typed costs from an endowment the governed policy cannot mint. Together these requirements define an economic constitution combining intrinsic path budgets, independent admission, protected audit, and empirical deployment gates. The mistake was not building agents that search for new paths. It was building that capacity without an intrinsic economy of its own consequences.
Conceptio Open Archive
Analyze & explore related research in Conceptio
Connect this record directly into your research workflow or AI agent. Explore citations across open repositories, query full text, and export structured proof bundles.
This document is indexed with metadata only: full text is not available in the archive for this record. Open the official source ↗
Related documents
- Objective-Layer AI: A Structural Definition of Endogenous Evaluation in Artificial Systems#99033
- Execution-bound advisory automation for agentic AI: a reproducible AIBOM-driven CSAF-VEX framework.#239950
- Specific Artificial Intelligence for Artificial Research by Deduction#71338
- The Governance Half-Life of Agentic Artificial Intelligence Controls: A Dynamic Assurance Framework for Control Decay, Residual Risk, and Revalidation#373137
- From Protocols to Evidence: Bounded Claims for AI in Service of the Common Good#1123230
Record · ID 701144
Retrieved via Conceptio: every document is proof-bundled with source, license, and retrieval metadata.
Retrieved via Conceptio: every document is proof-bundled with source, license, and retrieval metadata.
Bring this data into your agent: Connect via MCP or API · 20 free credits to start · no card required
Read the Docs → · View CLI source on GitHub · Conceptio Open Knowledge Archive
Read the Docs → · View CLI source on GitHub · Conceptio Open Knowledge Archive