The Phone on the Table Nobody Asks About Anymore

The Phone on the Table Nobody Asks About Anymore

A venture capitalist in San Francisco told me about something that’s become routine lately. When he walks into a meeting, founders set their phones on the table and slide them a little closer to him. A few years ago, they’d ask if it was okay to record. Now nobody asks.

This is the scene The Wall Street Journal recently described in Silicon Valley. Just a few years back, recording a meeting was something you handled carefully, given the security concerns. Now, not recording is what seems odd. What drove this shift wasn’t better recording technology — it was the reason people started recording in the first place. Once “for the meeting notes” became a legitimate business justification, recording turned into the default. Korea, for its part, went through this shift much earlier, shaped by its laws and its culture of recording phone calls.

The Meeting Room Where Asking Is Rude

The scenes The Wall Street Journal captured are strikingly specific. One startup founder wore a recording badge that lit up red at a work conference, and someone complained that it made them uncomfortable. He said the complaint itself was what bothered him. The same person records most of his dates too, and afterward asks an AI how much he talked and how he could show more empathy. Another AI startup CEO says that when people later realize a conversation in the office wasn’t recorded, everyone feels a pang of regret. And he added this: asking permission to record is what kills the mood.

The venture capitalist mentioned earlier described this situation as society turning into East Germany. But there are people on the other side too. One New York-based investor believes secretly recording conversations shouldn’t be socially acceptable, so he changed his displayed name on Zoom entirely — appending a line after his name stating that he does not consent to being transcribed or recorded. That someone would rather stamp this onto a name tag than verbally decline every single time says something about how burdensome refusal itself has become.

Standards vary state by state in the US. More than a dozen states, including California, require all-party consent to record private conversations¹. Apple didn’t add call-recording to the iPhone until 2024 — 17 years after the phone launched in 2007. Even in places where all-party consent is legally required, there are reports of people recording each other without disclosure.

The company leading this trend is Granola. Instead of a bot joining the meeting as a visible participant, it runs quietly on the user’s own computer, transcribing the conversation in the background. Because it doesn’t show up the way a bot joining a call does, other participants may not easily notice they’re being recorded. Founded in 2023, the company raised $125 million in March 2026 at a $1.5 billion valuation — six times the $250 million valuation it had just a year earlier. Revenue grew 250% in the most recent quarter alone. Established collaboration tools like Notion, Zoom, and Teams have all rolled out their own meeting-transcription features. What sets Granola apart is that it transcribes conversations from the user’s own computer instead of inviting a separate bot into the meeting. That said, it’s not a product designed to prevent disclosure. Its official documentation actually describes features like a chat message notifying participants of its use and a video watermark.

Korean Law Distinguishes Whether You Were Part of the Conversation

Reading this article, though, I found myself thinking about it a bit differently. For Korean readers, none of this is new. In Korea, it’s long been common for a participant in a conversation to record it themselves.

Korea’s Protection of Communications Secrets Act bans recording conversations between other people that aren’t public². The Supreme Court has ruled that when a participant in a conversation records it without the other party’s knowledge, this doesn’t violate that provision. That said, this ruling only concerns that specific clause — it doesn’t mean there are no restrictions whatsoever on disclosing, sharing, or using such recordings. On the other hand, secretly recording someone else’s conversation without being a participant carries a prison sentence of 1 to 10 years. The law is severe enough that there isn’t even a fine option — only imprisonment.

In Korea, transcripts made without the other party’s consent are routinely submitted as evidence in labor disputes. One labor attorney described this state of affairs as having turned the country into a “recording paradise.” In 2022, lawmakers proposed an amendment to criminalize recording without the other party’s consent, but it was withdrawn amid backlash. The pushback centered on the concern that without such recordings, proving wrongdoing without the other person’s consent could become much harder.

Of course, this freedom has its limits. In 2023, the Supreme Court convicted a civil servant who had recorded a conversation between his supervisor and a visitor in his office. He argued that it took place in a public-facing civil affairs office and that his goal was to catch evidence of wrongdoing, but the court didn’t accept this defense. Whether a conversation is one you participated in, or one held strictly between others, is what separates legality from a prison sentence. This is also precisely where AI note-taking apps get murky.

The same logic applies to devices. Domestic Galaxy users have long been familiar with a built-in call-recording feature. Apple only opened up this capability — in Korea and a handful of other countries — with iOS 18.1 in October 2024. But here’s the detail worth noting: when Apple’s recording feature is activated, it automatically plays an audio notice telling the other party that the call is being recorded — even though Korea’s Protection of Communications Secrets Act doesn’t require a participant to obtain the other party’s consent.

Even with an identical recording feature, how a product notifies the other party of its use varies. It’s worth paying attention to whether there’s an automatic notice, or whether the user has to disclose it separately. Naver’s Clova Note has become an essential app for many office workers, and the Korean startup Tiro has emerged as another notable player in this space.

The Moment Defense Became Convenience

The key reason covert recording gained acceptance in Korea was evidentiary need. It was the tool available to someone facing gapjil (abuse of power by a superior), sexual harassment, unfair directives, or unpaid wages to prove their claims. Defensive recording had three traits: it was exceptional, deliberate, and the moment you hit record, it introduced tension into the relationship. That’s why people were cautious about it.

AI note-taking apps didn’t change recording’s capability — they changed the motive behind it. Meeting minutes, action-item tracking, recovering context you missed. Recording stopped being an act of self-protection and became an act of doing your job well. Once people started using it for work convenience, some stopped deciding case by case and just left it running constantly. In the case featured in the article, it was actually the person who declined to be recorded who had to explain themselves.

This shift shows up in domestic user numbers too. Naver’s Clova Note had accumulated over 6.6 million registered users as of March 2026 — up roughly 24% in a single year. Naver Works Clova Note, its enterprise-focused product, launched in October 2024, and its paying customers have grown by more than 40% month-over-month on average since. SK Telecom’s A.Note passed 300,000 users within a week of launch. 6.6 million is more than a fifth of Korea’s roughly 29 million economically active population. That said, cumulative sign-ups and the economically active population are different groups — this doesn’t literally mean one in five office workers is using it.

How Granola describes itself has also changed. Heading into 2026, the company stopped calling itself a meeting-notes app and started calling itself an organization’s “context layer.” In February, it added MCP integration, letting whatever gets said in a meeting flow into other AI tools like Claude or ChatGPT³. According to Granola’s own documentation, de-identified data can be used to improve its service, and users can opt out in settings. Enterprise plans are excluded from training by default. The plan tier and the settings need to be considered separately.

Problems can also arise when you leave the room without turning off the recording. Say I step out for a moment — if the recording keeps running, whatever the people who stayed behind talk about isn’t a conversation I took part in. In legal terms, it edges much closer to being a conversation between other people. Turning it on no longer requires resolve, but that also means turning it off is easy to forget. A recording switched on for work convenience can end up capturing conversations it was never meant to.

Here’s the summary. What these apps give the user is a meeting summary. What accumulates on the service side is the entire conversation that took place inside the organization. In exchange for a summary, the user is effectively handing over the full conversation record. And that record includes what other people said, too. My consenting to use the app doesn’t mean the other person consented to having their own words handed over to the service.

Smart glasses are following the same path

Smart glasses are tracing this exact trajectory. Ray-Ban Meta and Oakley Meta launched officially in Korea on May 25, 2026. They start at ₩690,000 (~$497), and they’re sold not in electronics stores but in department stores and optical shops. Manufacturer EssilorLuxottica reported selling over 7 million AI glasses in 2025 alone — more than three times the 2 million combined units sold in 2023 and 2024.

Friction has grown alongside adoption. In May 2026, two cases of cheating using AI glasses were caught at TOEIC test centers in Korea — the first such incidents in the history of standardized language testing here. The fact that proctors had to catch it by eye tells you something about how these devices work. The BBC reported cases of men wearing the glasses to secretly film women in public and post the footage online. Meta is preparing an update, slated for July 2026, that would detect devices with tampered recording indicator lights and block filming.

And this is only the beginning of the rollout. Google is developing Android XR glasses in partnership with Samsung Electronics and Gentle Monster, and Apple has signaled it will enter the race too. As these companies’ products spread widely, everyday encounters with recording devices are likely to multiply. And because you don’t need to pull out a phone to record, it becomes harder for the other person to even notice.

Policy is moving in step with all this. On July 3, 2026, Korea’s Personal Information Protection Commission unveiled its third Basic Plan for Personal Information Protection, opening a path for legally collected voice and video data to be used for AI training in raw, unpseudonymized form, subject to a risk assessment⁴. The detailed criteria haven’t been finalized yet, so this shouldn’t be read as a settled conclusion. But the direction is clear. Users are handing over their voices themselves, and the system is moving toward expanding how those voices can be used.

Oswarld’s Lens

While building GTM strategies, I’ve watched the moment resistance to a new product collapses happen over and over again. Resistance doesn’t break down when the product gets better. It breaks down when a structure forms where not using it puts you at a disadvantage. AI note-taking apps are at that point now. You’re the only one without meeting minutes, the only one who can’t search back through what happened in last week’s meeting. The instant this asymmetry appears, individual privacy sensitivities lose their force. Norms don’t collapse because people turn bad — they collapse when the structure starts penalizing the people who keep them.

So does this mean we trust each other more now, or less? I’d say neither. It’s closer to trust becoming unnecessary. In the past, there was no good way to check later what someone had actually said, so you just had to take it on faith. Now, with recordings and automatic summaries, checking is easy — so the occasions requiring that leap of faith have simply shrunk.

The problem is that records can’t fully substitute for trust. What ends up on the record is only what gets said out loud, and the things that actually matter in an organization are often the things nobody says out loud — half-formed ideas, genuine dissent, early warnings about bad news. Because these kinds of remarks can come back to bite you once they’re on record, they tend not to surface while the recording is running. What worries me is the possibility that this kind of input migrates to hallway conversations that never get recorded at all. Even if meeting minutes become more accurate, if people stop voicing the opinions that matter in the room itself, the record alone won’t be enough to reconstruct how a decision actually got made.

That said, I want to be careful about the nostalgic story that we used to simply trust each other back in the day. That doesn’t square with the fact that Korea’s recording culture took root as a defense against gapjil (abuse of power by superiors) and sexual harassment. People didn’t trust each other back then either — the weaker party just had no evidence.

Closing

What broke the taboo on recording meetings wasn’t better technology — it was the justification of productivity. In Korea, the fact that recording by a participant in a conversation is legally permitted, combined with an already-familiar culture of recording phone calls, helped fuel its spread. And a significant portion of the data handed over for that convenience isn’t yours — it belongs to the person sitting across from you.

I’d suggest trying just one thing this week. Check your note-taking app’s settings to see whether data training is turned on. Defaults vary by service and pricing tier, so take a moment to check what material is being used for training, and by whom.

Which side are you on? If you’re the type who turns on an AI note app in meetings, do you tell the other person before turning it on — or, if you’ve been on the other side, how did it feel to find out the other person had turned theirs on? I’m especially curious about any moment you hesitated over whether to turn it on — tell me about it in the comments.

---

📨 If you have a colleague who sits through a lot of meetings, pass this along to them.

---

References & Further Reading

Primary sources

• The Wall Street Journal, “This Conversation Is Being Recorded. They All Are.”, 2026. Link ··· This is where today’s piece started. The scene of someone recording a blind date to get AI feedback on it is the clincher.

• Seoul Economic TV (a Korean business news channel), “Clova Note’s Paid Subscribers Up 40%, AI Meeting-Notes Revenue Hits Its Stride”, March 31, 2026. Link ··· The source for Clova Note’s cumulative 6.6 million subscribers and its enterprise growth rate. It gives a sense of the scale of the Korean market.

• TechCrunch, “Granola raises $125M, hits $1.5B valuation as it expands from meeting notetaker to enterprise AI app”, March 25, 2026. Link ··· This piece lays out the shift from meeting-notes app to context layer.

• Personal Information Protection Commission (PIPC), Korea’s data-protection regulator, “The 3rd Basic Plan for Personal Data Protection to Promote Trust-Based AI Innovation (2027–2029)”, July 3, 2026. Link ··· This contains the direction for using raw voice and video data in AI training. The detailed implementation standards haven’t been released yet, so read it with that caveat in mind.

• Granola, Model training policy ··· This distinguishes between data use for improving the service itself and the default opt-out for Enterprise accounts.

• Granola, Participant notice feature ··· A guide to the feature that discloses recording via chat messages and video watermarks.

• Supreme Court of Korea, Case No. 2006Do4981 ··· A ruling on whether a conversation participant’s own recording counts as recording “a conversation between others” under Article 3, Paragraph 1 of the Protection of Communications Secrets Act.

Background

• Kwangseon Lee, “The Employee Who Records Every Conversation and Call — Can the Company Discipline Them?”, Korea Economic Daily’s CHO Insight, November 2023. Link ··· A practitioner’s-eye view of why Korea became a recording paradise, and where the legal boundaries of self-recording lie.
• Edaily, “iPhone Call Recording Coming This Month — Supported on iPhone XS and Later”, October 2024. Link ··· Explains why Apple added a recording-notice feature in Korea even though it’s not legally required there.
• Platum (a Korean tech media outlet), “Smart Glasses Have Finally Become Invisible”, July 2026. Link ··· Covers the domestic launch and even a case of smart glasses being caught at a TOEIC exam hall — the whole arc of the eyewear trend in one piece.

Past issues worth reading alongside this one

• The People Who Switch Off Their 20-Watt Brains ··· This covered what changes when we hand judgment over to AI. If today’s piece is about outsourcing memory, that one is about outsourcing thought.
• Why Does the Rooftop-Room Developer Deliver Food Every Night? ··· A look, from the labor side, at how the excuse of “convenience” reshapes an individual’s conditions.

---

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. All-party consent: The principle that recording a private conversation requires the consent of every participant. In the US, the standard varies by state — roughly a dozen states, including California, follow this rule, while the rest only require consent from one party. ↩

2. Article 3 of the Protection of Communications Secrets Act: This clause bans recording or listening in on undisclosed conversations between other people. The key phrase here is “between other people” — a conversation you yourself took part in doesn’t fall under it. That said, whether a specific case is unlawful depends on the circumstances, so any real dispute needs expert consultation. ↩

3. MCP (Model Context Protocol): A standardized specification that lets AI tools connect to external data or other services. Once a meeting-notes app plugs this in, all the conversation history it has accumulated has a channel to flow into other AI services. ↩

4. Pseudonymization: The process of removing or altering identifying information like names and contact details so that no one can be identified without additional data. Until now, this was, in principle, a required step before personal data could be used for AI training. ↩