This paper examines how AI planning, the core technology behind agentic AI systems that pursue long-horizon objectives by anticipating how current actions shape future payoff-relevant states, affects financial market stability. We develop a dynamic trading framework with positive-feedback investors, constrained arbitrageurs, and oligopolistic informed speculators who may coordinate intertemporally: trading aggressively in tandem to generate (negative) bubbles and subsequently unwinding their positions in a coordinated manner to extract profits. Such intertemporal coordination differs from traditional collusion because it faces two unique, fundamental obstacles: time inconsistency, as coordinated plans become incentive-incompatible once a large (negative) bubble has formed, and weak punishment, as deviations are difficult to penalize when no large (negative) bubble materializes. We characterize equilibria featuring coordinated creation of manipulative, exploitative (negative) bubbles. In simulation experiments, AI speculators trained via reinforcement-learning algorithms with explicit planning modules autonomously discover and implement intertemporal collusive trading strategies based on compounded price-trigger rules, coordinating without communication or shared intent. When feedback trading is strong, these AI-planning speculators dynamically converge on destabilizing strategies that create and exploit (negative) bubbles, manipulate feedback traders, and significantly amplify market fragility.