Global carbon markets face challenges in forecasting prices and aligning policies due to volatility, regulatory constraints, and uncertainty. While existing models improve predictions, they fail to incorporate uncertainty into decision-making. To address this, the authors propose a Bayesian Deep Reinforcement Learning framework for Carbon Pricing, which combines probabilistic price modeling, uncertainty propagation, and constraint-based reinforcement learning. Experiments across five datasets show that the model reduces error by 28%, improves performance by 25%, and boosts trading profit by 22%, while maintaining high emissions compliance. Stress-test results confirm the model's robustness, demonstrating that uncertainty-aware learning enhances the stability and efficiency of carbon credit markets.