wut does this AI tool actually do — and wut’s the catch?

A bot that joins your video calls, records them, and writes up the notes.

The catch: The legal risk lands on you, not on Otter. A consolidated federal class action alleges the Notetaker recorded and transcribed meeting participants without their consent and used the content to help train Otter's models. Otter denies it and has moved to dismiss, and no court has ruled either way. But in all-party-consent states like California, the person who brought the bot into the meeting is the one exposed — and that person is you.

Another bot that joins your meetings, records them, and writes the notes.

The catch: A different legal theory from the one aimed at Otter, and a more uncomfortable one. Two Illinois suits allege Fireflies did not merely record people — it turned their voices into biometric identifiers and kept them, without the written consent Illinois law requires. One plaintiff says she never had an account and never agreed to anything. Fireflies has not been found liable, and the first case was voluntarily dismissed.

A meeting assistant that joins, transcribes and summarises your calls — and spreads by doing it.

The catch: Its growth mechanic is the problem. Read AI can attach to a calendar and join, record and summarise meetings even when the person who installed it is not in attendance. Chapman University investigated after repeated reports of the bot turning up uninvited, and in August 2025 prohibited it outright over security, privacy and institutional-data risk. That is not a lawsuit — it is an organisation looking at how the product spreads and deciding it is not safe to allow.

The AI-writing detector built into the plagiarism checker your school already pays for.

The catch: The famous "under 1% false positive" figure is doing a lot of work. Turnitin's own chief product officer has said the system deliberately misses up to 15% of AI-written text in order to hold that number down — and Vanderbilt pointed out that 1% of 75,000 papers is still around 750 wrongly accused students a year. Independent research also finds the errors are not evenly spread: detectors misclassify non-native English writing at dramatically higher rates.

The AI detector teachers reach for when the school has not bought one.

The catch: The gap between the marketing and the measurements. Vendor accuracy claims in this category run 87–99%, while independent testing has put GPTZero's false-positive rate as high as 12% in one comparison and around 23% on real student essays in classroom testing — one analysis estimated a teacher relying on it would falsely accuse roughly one in five innocent students.

An AI detector sold to publishers and agencies for screening writers, not to schools for screening students.

The catch: The buyer and the judged are different people, and only one of them ever sees the score. It is built to screen freelance submissions at volume, so a false positive does not start an academic hearing — it quietly ends a working relationship. Writers report losing clients over flagged work, and human articles written years before ChatGPT existed have come back scored as AI.

Point your phone at a plate of food and it tells you the calories.

The catch: Two catches. The indie underdog story is over — Cal AI was quietly bought by MyFitnessPal, the incumbent it was beating, in a deal that closed in December 2025 and was not announced until March 2026. And the core feature measurably undercounts: an NIH study presented in July 2026 found photo-based calorie apps, Cal AI among them, came in 250–345 calories per meal too low.

A very good AI chatbot that is free, fast, and Chinese-owned.

The catch: Not the model — the app. DeepSeek's own privacy policy states plainly that it collects, processes and stores your personal data in the People's Republic of China, and that training on your data is the default you must opt out of. Italy's data protection authority blocked it outright in January 2025. But there is a real escape hatch almost nobody mentions: the model weights are open, so running DeepSeek somewhere else removes the entire problem.