Methods based on Thompson Sampling for safe linear bandits that significantly improve computational costs while retaining regret and risk performance.