Anthropic began asking Claude users on 4 October 2026 whether it may use their voice conversations to improve its AI models, BleepingComputer reported. The prompt reads “Allow us to use your voice data to improve our AI models”, and the setting is off unless the user turns it on [1].
Background
Voice conversations with an assistant produce recordings or transcripts that can carry a person’s voice, surroundings and personal details. Whether a provider uses them for training is governed by privacy settings that differ between products and between types of data.
What the report describes
Users see two choices, “Allow” and “Not now”. A separate toggle sits in Claude’s Privacy settings, is disabled by default and can be switched off at any time; voice data can be deleted through the settings [1].
The voice toggle is independent of Anthropic’s existing programme that uses chats and coding sessions for model improvement, so a user can hold a different preference for each type of data [1]. The report also notes that some readers reacted with scepticism in the comments [1].
The report also records the date of the prompt, 4 October 2026, and describes it as a voluntary request. A user who dismisses it with “Not now” keeps the default, and a user who accepts can reverse the decision later in the same Privacy settings [1].
Implications for Mac users
Anyone who uses Claude by voice and sees the prompt faces a choice that, according to the report, defaults to no. The report does not say how long accepted voice data is retained, whether it is reviewed by people, or how the setting appears in every client, including the desktop app. Those details would be in Anthropic’s own documentation, which the report does not quote. Because voice recordings can carry a person’s voice and the sounds around them, users who share a room or a device with others may also want to consider whether those people have agreed to a recording being used for training.
Recommendations
- Open the Privacy settings in Claude and check the voice data toggle and the separate chat and coding toggle.
- Choose “Not now” if there is no need to contribute voice data, since the setting is off by default.
- Delete voice data through the settings if it should no longer be held.
- Review the same kind of setting in every other AI assistant in use, as each product has its own.
Relevance to FireAI
FireAI sends no telemetry from a Mac, and its own on-device AI model runs only on the Mac after a one-time download. FireAI does not control what an AI service does with data a user chooses to give it: it cannot read the content of an encrypted connection and does not change an app’s privacy settings. What it does offer is a view of which apps reach the internet, and per-app rules that let a user block an app’s connections.
Limitations
The only source here is BleepingComputer’s summary of the prompt and settings [1]. Retention periods, human review, regional availability and whether the prompt covers every Claude client are not stated.
To see which apps connect from a Mac, download FireAI and try it free for 17 days. FireAI is made by HisnLabs.