The Plaud AI voice recorder, a small plastic rectangle the size of a credit card, is currently being talked about more than some government ministries. Cleaners found one at the end of January in the meeting room of the Czech government, and instead of anyone handing it over to the security services it ended up in a drawer in the secretariat of the head of the Office of the Government, where nobody checked it for months. After a story on Seznam Zpr\u00e1vy [1] (picked up by Novinky.cz [2] and other outlets), the minister for sport Boris \u0160\u0165astn\u00fd claimed it, saying it was his personal note-taking aid which he had bought in a shop and lost. The internet promptly nicknamed the affair “Plaudgate”. Let us look at it non-politically, practically, and from the point of view of someone who has worked with AI tools for ten years.<\/p>\n\n\n\n
The whole story is both funny and unsettling. But above all it is a good excuse to answer, without prejudice, a question a lot of people do not ask until they have bought a device like this: is Plaud safe?<\/strong> And what does it even mean to own an AI device that listens to what you say?<\/p>\n\n\n\n What it actually is<\/strong><\/p>\n\n\n\n Plaud is a small voice recorder with artificial intelligence. It comes in several versions; a popular one attaches to the back of your phone with a magnetic case, you press the button and it simply records. A meeting, a phone call, a lecture, anything. The difference from a classic 1990s dictaphone lies in what happens to the recording afterwards: the app uses artificial intelligence to transcribe it into text automatically and produces a summary, a task list or a tidy report from it. For people who spend hours in meetings and then more hours writing up the minutes, that is a tempting proposition, and the device really is selling in the millions.<\/p>\n\n\n\n Why people love it so much<\/strong><\/p>\n\n\n\n I have been using Plaud myself since around the spring of 2025 and as an aid for transcribing meetings, lectures or quick notes on the move it is excellent. It will record all day on a single charge, it saves a great deal of the time I would otherwise spend writing up minutes by hand, and the summaries the AI produces are usually usable straight away. That is precisely why the device has spread so widely among people who have a lot of meetings and little time for paperwork.<\/p>\n\n\n\n Where what you record gets sent<\/strong><\/p>\n\n\n\n Here we have to separate two things. The audio recording itself is first stored on the device and then on the phone, so that stays local and goes nowhere. But as soon as you want the recording transcribed or summarised, it has to be sent to a server, to the cloud<\/strong>, where AI-based tools do the job for you. There is no other way: ordinary phones, and certainly not a tiny chip in a recorder, do not yet have the power to handle a transcription like that on their own.<\/p>\n\n\n\n And that journey “outwards” is the heart of the whole debate. According to the manufacturer’s own statements the recording is deleted from the server after processing, and the companies that do the transcription (large AI companies such as OpenAI, Google or Microsoft) are contractually forbidden from using the data for anything else. The question, though, is who the data goes to and where it travels in the meantime. Which brings us to why people worry about Plaud (and about some other AI recording devices too).<\/p>\n\n\n\n Your voice around the world<\/strong><\/p>\n\n\n\n The company behind Plaud has a fairly tangled structure. The app development and part of the management sit in San Francisco in the USA,<\/strong> but the hardware is made in Shenzhen in China<\/strong> and the founder of the company is a Chinese citizen<\/strong>. On top of that, according to the terms and conditions, European customers do not enter into a contract with the American branch but with a company registered in Hong Kong.<\/strong> In itself that would not have to matter (plenty of electronics is made in China), except that in this case it goes hand in hand with data being sent for processing through a structure that is not entirely transparent.<\/p>\n\n\n\n In practice that means that when you use Plaud in Prague, your voice may end up on a server in the USA,<\/strong> pass through a contractual structure running via Hong Kong,<\/strong> and two different sets of rules hang over the whole chain: American law,<\/strong> which under certain circumstances allows the authorities there to request access to data held by American companies,<\/strong> and Chinese data security law, <\/strong>which gives similarly broad powers to Chinese authorities over companies with ties to China.<\/strong> Neither of those automatically means somebody is listening to you, but it does mean your recording passes through several jurisdictions at once, which is exactly what security experts and corporate IT departments abroad look at when they judge whether to let a tool into a sensitive environment.<\/p>\n\n\n\n Last autumn the Taiwanese authorities noticed as well, raising security objections because of the founder’s ties to China, and the company responded only with a public statement. Similarly, a number of European companies and hospitals prefer not to use Plaud in more sensitive settings. Not because they have evidence of data misuse, but because the risk simply cannot be fully ruled out or verified.<\/strong><\/p>\n\n\n\n The second risk: hallucinations<\/strong><\/p>\n\n\n\n Where the data goes is only half the story, though. The second risk has nothing to do with China or the cloud. It is a technical matter: how accurate the transcript actually is. Plaud does not have a professional studio microphone of the kind journalists, reporters, podcasters, court stenographers and professional transcribers use. It is a small plastic tile that has to cope with the echo in the room, with people talking over each other, with throat-clearing, background noise, speech impediments, car horns outside the window, coughing, or someone sitting further from the microphone. When the recording cannot cope with all that, the result is not merely an unintelligible passage in the audio.<\/p>\n\n\n\n