Sunday, 22 November 2026

TikTok’s Shou Chew warns AI progress may outpace our safety plans

The TikTok chief expects a real scientific breakthrough from AI within three years, and says the security risks of AI agents are already here.

TikTok
Photo: Nik Kranjec / CC BY-SA 4.0 (modified)

The short version

  • TikTok chief executive Shou Chew said AI could produce a real scientific breakthrough within two to three years.
  • Speaking at a DBS conference in Singapore on Oct. 12, he pointed to chemistry, physics, biotech and materials science.
  • Chew called AI cybersecurity risks "quite real" and said people hand AI agents permissions they don't fully understand.
  • He also said the world is probably overestimating AI in the short term and underestimating it in the long term.
Quick read · 2 min

TikTok chief executive Shou Chew thinks AI will deliver a real scientific breakthrough within two to three years. Speaking at a DBS conference in Singapore on Oct. 12, he pointed to physics, biotech, chemistry and materials science.

He paired that with a warning. Chew called the cybersecurity risks of AI "quite real" and said people are giving AI agents permissions they don't fully understand. An agent allowed to shop with your card, he said, could buy much more than you need.

  • He believes AI is overestimated in the short term and underestimated in the long term.
  • He said stronger models make spam and attacks on platforms far cheaper.
  • He urged people to try AI tools themselves to learn their limits.

TikTok's parent, ByteDance, is building its own science models, including one for predicting the structure of biological molecules, and Chew said it wants to work with scientists.

Give an AI agent your credit card, and it might buy far more than you wanted. That was the everyday example TikTok chief executive Shou Chew picked when he was asked which part of artificial intelligence gets too little attention. In the same conversation he made a much bigger call: AI science breakthroughs in fields like physics and biotech are, in his view, two to three years away.

Both halves matter to ordinary people. The optimistic half points to new medicines and materials arriving sooner. The cautious half is about AI agents, the kind of helper that doesn’t just answer questions but carries out tasks such as shopping on our behalf. Chew’s message was that people are handing these agents permissions without really understanding them.

01

What Shou Chew said about AI science breakthroughs

Chew said he expects AI to deliver a real scientific breakthrough within the next two to three years. He listed four fields where it could land: chemistry, physics, materials science and biotech. He described that outcome as “very, very, very possible.”

He was speaking on Monday, Oct. 12, in Singapore, at the DBS Asia Future Forward Conference. DBS, the Singapore bank, set up the event as a new flagship gathering with AI at the center of the agenda, and it expected about 700 of its corporate and wealth clients to attend. Intel chief Lip-Bu Tan and Broadcom chief Hock Tan were also on the speaker list.

Chew pointed to the progress AI has already helped make in science as his reason for the forecast.

Shou Chew
Photo: Photographer: Lukasz Kobus / CC BY 4.0 (modified)
02

Why he thinks AI is overhyped now and underrated later

Chew’s view is that the excitement is ahead of the technology today, and behind it over the long run. Asked about the extreme reactions AI draws from fans and critics, he said the world is probably overestimating it in the short term and underestimating it in the long term.

His reasoning for the slow start is about people, not machines. Humans need time to adopt a technology like this, he said, and that friction, which he called “sand in the gears,” is likely to hold progress back for a while. He warned against reading too much into that. A slow start, he argued, doesn’t mean the technology won’t become very important later.

He also said AI still falls short of the human brain in some ways. People remain better at the fine points of visual recognition and at pulling up long-term memories, and models still can’t learn certain things as quickly as a person can.

On whether AI could replace people, Chew said it was too early to say. He did allow that machines built on silicon could one day beat humans at some things, noting that silicon lasts longer, scales more easily and connects information more simply than people do. He put it as a “non-zero chance.”

03

The AI agent risk he says is already real

The risk Chew stressed most was cybersecurity, which he called “quite real.” His concern is with AI agents. Chatbots mainly answer questions. Agents are built to carry out tasks, which means they need permission to act.

Chew said users may be granting those permissions without really understanding what they are agreeing to. His shopping example makes the point: someone who lets an agent use their card to make purchases could end up with much more than they need, because the agent is so eager to finish the job.

Companies face a different version of the problem. Chew said stronger AI models have sharply cut the cost of producing spam and of attacking platforms such as TikTok, which had about 2.21 billion users worldwide according to an April estimate from research firm DataReportal. Spam can never be cut to zero, he said, so the goal for a business is to make spamming too costly to be worth it.

He added a workplace warning too. Staff need to be trained not to paste trade secrets or privileged legal material into public chatbots, he said, because people can get overenthusiastic with a tool this useful.

Home desk with AI-assisted device and soft glow.
04

What this means if you use AI tools

For everyday users, Chew’s advice came down to two things: know what you are allowing, and get hands-on. He said he sees a need for “a lot of education” on managing AI risks, and that learning to use AI well will be an important skill over the next two to three years.

Reading about these tools isn’t enough, in his view. He said people who don’t try the models and products for themselves can’t work out what the tools can and can’t do.

His warning was specific. Chew said the security risks are real in the short run, that businesses need the right tools to defend themselves, that people need a better understanding of what they are permitting, and that knowing the limits of reasonable use matters.

05

Where TikTok and ByteDance fit in

Chew has a stake in both sides of this. TikTok’s parent company, ByteDance, is an AI developer in its own right. Its research group, called Seed, released a tool named Protenix-v2 in April that predicts and designs the structures of biological molecules. ByteDance also runs Anew Labs, a unit that uses AI to search for new drugs. Chew said the company is looking at how it can work with scientists.

The wider industry is heading the same way. The week before his talk, the US government, Google and Meta joined an AI biology effort worth $1.8 billion that aims to model how cells behave.

On TikTok itself, AI video is everywhere, but Chew said viewing habits tell a different story. A June report from video tool maker Kapwing found that about 59 percent of the videos shown to a brand-new account’s For You feed were low-quality AI content. Chew said people who only watch AI content make up a low, single-digit share of users, and that viewers still push back on videos that are obviously not real.

06

Quick answers

When does Shou Chew expect an AI scientific breakthrough?

Within the next two to three years. The fields he listed were chemistry, physics, biotech and materials science.

What AI risk did the TikTok CEO warn about?

Cybersecurity. He said people give AI agents permissions without understanding them, and that stronger models make spam and platform attacks much cheaper.

Does Chew think AI will replace humans?

He said it is too early to be sure. He believes there is a “non-zero chance” AI becomes superior to people in certain respects.

You're reading the quick version.