To address the shortcomings of existing swarm intelligence optimization algorithms, such as reliance on positive feedback mechanisms, susceptibility to premature convergence, and insufficient adaptability to complex multimodal problems, this paper proposes a novel swarm intelligence optimization algorithm—the Atlantic Poison Dart Frog Optimization (APDFO). Inspired by the "toxicity rejection" ecological behavior of the Atlantic poison dart frog, this algorithm introduces toxicity deposition, risk accumulation, and search space pruning mechanisms, modeling the optimization process as a self-reconstruction process of the search space dominated by negative evidence. Furthermore, at the algorithmic level, this paper proposes irreversible risk topology, toxicity causal memory, toxicity-induced phase transition migration, and an anti-optimal attraction mechanism, and theoretically abstracts a new optimization paradigm—Self-Destructive Optimization (SDO). This paradigm no longer relies on the continuous reinforcement of the optimal solution, but instead promotes the search to migrate to unknown but potentially valuable regions by systematically excluding low-value regions. This paper systematically elucidates the algorithm's structure and dynamic mechanism from a mathematical modeling perspective, providing a new theoretical and methodological framework for complex black-box optimization problems.