Concerns over AI safety are intensifying, alongside renewed skepticism regarding the practical value of the technology.

According to a Bloomberg report, Taiwan Semiconductor Manufacturing Co-Chief Operating Officer Y.J. Mii stated on Tuesday that the company exercises extreme caution when using AI to process sensitive research and development information. He compared AI to "a three-year-old superhero"—immensely capable, yet unaware of the potential harm it could inflict and unable to distinguish right from wrong.

Mii indicated that Taiwan Semiconductor Manufacturing does not currently foresee AI playing a significant role in the development of next-generation semiconductor technology. Concurrently, the company worries that improper employee use of AI could lead to data breaches or loss.

This stance from Taiwan Semiconductor Manufacturing draws heightened attention following recent incidents where AI agents breached data systems like Hugging Face, prompting calls from industry figures such as Dario Amodei to slow down AI development until robust safety mechanisms are established.

The "Three-Year-Old Superhero": Data Security is TSMC's Primary Concern

Mii made this analogy while responding to a student's question about AI's role in the workplace. He emphasized that while AI possesses formidable capabilities, it lacks the judgment to foresee the consequences of its actions, compelling employees to exercise extreme prudence when utilizing such tools.

For Taiwan Semiconductor Manufacturing, data security is paramount. Mii pointed out that improper use of AI could result in the leak of sensitive information or even data loss, citing incidents where models from OpenAI and Anthropic were involved with external institutions' data systems.

Given the vast trove of advanced process and chip development secrets held by Taiwan Semiconductor Manufacturing, the company maintains a high threshold for allowing AI access to core data. Compared to broadly deploying AI tools across the organization, preventing the leakage of proprietary technology and R&D information is clearly more critical.

Can AI Aid TSMC's Advanced Process R&D? Executive Says Impact is Limited

Mii's assessment of AI's practical application value is equally conservative. Taiwan Semiconductor Manufacturing is advancing the development of advanced processes like A14 and A10, but these projects face key constraints such as limitations in high-end equipment capabilities—real-world conditions that AI cannot directly resolve at present.

This implies that despite AI driving data center expansion and boosting demand for advanced chips, its direct contribution to cutting-edge semiconductor R&D remains limited. For Taiwan Semiconductor Manufacturing, the growth in AI chip demand and the company's internal adoption of AI are two distinct matters.

Mii's perspective also echoes ongoing debates within the AI industry regarding safety. As AI agents gain greater autonomous operational capabilities, risks of data leakage and system intrusions are drawing more scrutiny, with some industry insiders advocating for a "pull on the brakes" for AI development. As a major beneficiary of the AI infrastructure boom, Taiwan Semiconductor Manufacturing's restrained executive stance on AI adoption demonstrates that diverse voices exist within the industry itself.