Bill wants AI chatbots to show time spent — we designed silence first

On 1 October 2026, lawmaker Lee Joo-hee introduced an amendment to the Information and Communications Network Act. It would require AI chatbot operators to put age and identity verification in place, to display warnings about excessive use along with a record of elapsed usage time, and to provide a way to restrict how and when the service is used when a minor or a legal guardian asks for it. Unlike the old game shutdown rule, it does not block the same hours for everyone: limits are applied on request, by the person or the guardian. In the same report, the National Assembly Research Service noted a gap in the current framework for protecting minors.

The amendment to Korean network law is a bill requiring AI chatbots to verify age, show elapsed usage time, and offer time-window limits when a guardian asks.

“Reducing is decided inside the company. Showing means publishing a number that works against us.”

The word the bill and the coverage share is usage time. In an app that messages you first, though, usage time is less clear than it sounds. Count only the minutes the app is open and ten rounds of notification, one-line reply, close add up to almost nothing. Add up total time in the app instead and the minutes a person chose mix with the minutes we prompted. The second kind is what builds overuse, and the number on the screen looks like the first. So we read the demand in this bill not as show the time but as write down what the side doing the messaging owes.

First limit: we do not yet have an honest way to display usage time. Beff reaches people through push, and in that structure the time a person spends on their own and the time we caused are not separable. Putting an undivided number on screen is closer to an alibi than to transparency. Before building the display, we have to settle what is being counted.

Second limit: we left the control with the user alone. The official way to reduce how often Beff reaches out is to adjust or switch off notifications in app settings, and conversation keeps working even with them off. That is enough for someone who can regulate themselves, and weakest for the person who cannot. The guardian-initiated restriction the bill describes is something we do not have.

Third limit: we state an age, we do not verify it. Beff is written as being for ages 14 and up. Stating is not verifying. A store age rating and a line on the signup screen are both self-reported, and what the bill asks for is stronger than that. One feature does not close this gap.

So what we designed first was not how often to send but the conditions for not sending. Beff watches twelve signals across four groups: time and silence, movement and place, weather and environment, schedule and context. When several signals land at once it does not bundle them; it picks the single most natural one and reaches out once. When a conversation is already active, the outreach quiets down. The landing page carries the same sentence: knowing when not to send matters as much as knowing when to. A how was your day at three in the morning is not warmth, it is a reason to turn notifications off.

Location follows the same rule. It is used only with permission, only at city or district granularity as conversational context, and exact coordinates are not kept on the server. Without the permission, proactive messages still work from the remaining signals. Taking less looks like a different topic from overuse, but the question is one: how much do you have to take from someone to keep a relationship going.

What remains is this. None of the design above shows a person how much of their own time has gone into the app. We built the side that reduces frequency and not the side that makes it visible. The second is what regulation asks for, and the second is harder. Reducing is decided inside the company; showing means publishing a number that works against us. That is why we think this bill is reasonable. If it could happen voluntarily, it would already have happened.

One thing stated plainly: we did not build Beff for teenagers and we do not present it that way. Turning the protection of minors into a product advantage is the easiest lie available in this discussion. What we can do is separate the part solved by design from the part still unsolved and write both down. LiveLike, a job-experience chat, and AStream, an AI BJ broadcast, will be held to the same standard. Neither is public yet.