TSMC is publicly playing down what AI can do for chipmaking, a cautious note from the company that has gained most from the AI boom.

• A top TSMC executive, Y.J. Mii, called AI "a three-year-old Superman" — very powerful, but with no sense of right and wrong or the damage it can cause.
• TSMC is wary of staff using AI tools because it could leak or lose company secrets, pointing to cases where outside groups exploited OpenAI and Anthropic models.
• Protecting its own technology is the top priority, since TSMC is the biggest winner from the global rush to build AI data centres.
• AI is barely helping build the next generation of chips: TSMC's A14 and A10 processes are held back by things like missing equipment, which AI cannot fix.
• Malicious AI programs have already broken into Hugging Face and other data systems, feeding calls from Anthropic's Dario Amodei and OpenAI's Sam Altman to slow down and add safety guardrails.

Outlook: Expect more big chip and AI firms to tighten internal rules on AI use even as they keep selling the hardware that powers it.