Users in dark, Tech giants in control

Is your data feeding AI behind your back?

Bengaluru: India may be courting AI innovation, but a data-privacy storm raging in the United States is a cautionary tale the world cannot ignore. As Silicon Valley scrambles to build ever-smarter AI models, millions of Americans are discovering an unsettling reality — their personal data is becoming training fuel for AI systems, often by default, and sometimes without clear consent.

The debate now gripping the digital world asks a direct question: Do users really own their data when tech giants own the AI built from it? The AI race in the US has transformed from a competition of features to a competition for data access, leaving users trying to decode privacy policies they never signed up to audit.

Companies including Meta, Google, and LinkedIn have launched AI functionality capable of pulling information from public profiles, emails, attachments, browsing activity, call logs, and AI chat prompts depending on account settings and permissions.

Some platforms provide opt-out options, others do not — a patchwork of controls experts say underscores America’s lack of federal AI-privacy regulation. Krystyna Sikora, research analyst at the German Marshall Fund’s Alliance for Securing Democracy, told Al Jazeera that tech firms remain largely non-transparent about what data is collected and how it is repurposed, creating conditions ripe for online panic and misinformation.

Without a national law, there is no uniform legal right for US users to opt out of AI training, leaving control entirely at the discretion of platform policies or lawsuits.

The biggest claims gaining traction online include:
A November 8 Instagram post warning, “Gmail just flipped a dangerous switch on October 10, 2025 and 99% of Gmail users have no idea.”
A November 9 video on X, viewed over 1.6 million times, alleging, “Every conversation, every photo, every voice message fed into AI and used for profit.” While not all claims were accurate, the underlying anxiety is rooted in partial truths embedded in sprawling privacy disclosures.

Meta scrapes public posts, no opt-out A viral November 9 post on X alleged that Meta would read all DMs starting December 16 and funnel every message, image, voice note into AI training for profit. Meta confirmed a new policy does take effect December 16, but clarified that this policy alone does not send direct messages, voice messages, or private media into its AI training datasets.

Instead, the policy explains how the company will customise content and advertising based on user interaction with Meta AI. The company, according to Al Jazeera, explained that when someone queries Meta AI about a topic like hiking, its systems may later show personalised ads or recommendations related to that topic. However, Meta made a key distinction:
Instagram, WhatsApp, and Messenger private messages are not used to train its AI models.
Content set to ‘public’ visibility can be collected for AI training. This includes posts, photos, comments, reels, and reels’ derivatives such as Reels or public captions.
The AI voice feature accesses microphones only when permission is granted.

Yet, Meta included a serious caveat: AI systems may train on data about non-users if their details appear in someone else’s public post or caption. For example, a Meta user tagging a friend without a Meta account in a public caption may expose that non-user to AI data ingestion.

Asked whether users can block this extraction, Meta’s spokesperson said no platformwide opt-out exists. Even deleting accounts does not remove past public data already captured. WhatsApp users can only disable AI chat one conversation at a time, forcing a repetitive manual opt-out process, chat by chat.

The viral post advised users to submit a Meta opt-out form, but the form does not stop AI training — it only lets users report if Meta AI responses contain someone’s personal information. David Evan Harris, AI ethics scholar at UC Berkeley, told Al Jazeera that in countries like Switzerland, the UK, and South Korea, users enjoy legal opt-out rights for AI training — rights US citizens do not have under federal law. He also warned that even when opt-out tools exist, settings are often hidden and difficult to locate.

Google AI access: opt-in vs defaults
Google launched Gemini Deep Research on November 5, enabling it to connect to Gmail, Drive, and Chat — but only after explicit user permission, as reported by Forbes. If a user enables AI, they may choose what sources the AI accesses including Google Search, Gmail, Drive, or Chat.

Beyond Gemini Deep Research, Google disclosed other data-collection pipelines for AI interaction:
Searches and prompts in Gemini apps on mobile or web browsers including Gemini inside Chrome or other web browsers.
Photos and videos uploaded by users into Gemini.
YouTube and Spotify data if permission is granted.
Call logs and message logs if permission is enabled.
Gmail and Google Workplace Smart Features, which are defaulton in the US, allowing Google to suggest calendar events or help compose emails.
Paid subscriptions unlock additional AI features including Gemini summaries.

Google confirmed it does not train AI on data from users under age 13, adhering to data restrictions around children. However, a California lawsuit alleges Google’s October 2025 policy update changed Gemini access from opt-in to default-on, requiring users to disable access manually in settings.

The suit claims this violates the 1967 California Invasion of Privacy Act, prohibiting unauthorised recording or interception of confidential communication without explicit consent. Legal experts say this case may shape how courts interpret AI harvesting without national regulation.

Google said privacy conservative users can use temporary chats or log out of Gemini, stopping history storage, though it does not fully block access to the Gemini app itself in browsers unless Smart Features are disabled.

LinkedIn trains on profiles, opt-out exists
LinkedIn confirmed that starting November 3, it will use US members’ profiles and public posts to train its content-generating AI models, noting private messages are not used. The data collected feeds AI improvement and also flows to Microsoft affiliates for personalised ad targeting, including profile data, feed activity, and ad engagement.

Unlike Meta, LinkedIn allows users to opt out. To block AI training, users can disable ‘Data for Generative AI Improvement’, stopping the platform from using their content to improve generative AI responses. Users can also disable personalised ads by turning off affiliate data sharing and partner sharing in advertising settings.