The increasing use of artificial intelligence (AI) in knowledge-intensive work raises questions about how it affects employees' sense of creative capability. Some perspectives view AI as scaffolding that augments creativity, while others highlight its potential to undermine agency and ownership. We address this debate by examining how AI use relates to creative self-efficacy through task-level attributions of control. Data were collected on 389 tasks reported by 58 engineers in a leading global semiconductor company. For each task, employees rated their performance, creativity demands, AI use, locus of control, and creative self-efficacy. Multilevel analyses revealed that AI use amplified attributional patterns. When performance was high, particularly in creativity-intensive tasks, employees were more likely to attribute control to themselves and reported higher creative self-efficacy. When performance was low, they were more likely to attribute control to the AI and reported lower creative self-efficacy. These findings suggest that AI does not simply empower or disempower employees but acts as an attributional amplifier, magnifying self-serving biases that shape beliefs about creative capability. The study extends attribution and self-efficacy theories to AI-supported work and highlights practical risks of overconfidence after AI-assisted successes and reduced self-belief after failures.