Hyderabad: With competition intensifying between the tech companies, the landscape of Artificial Intelligence is now being pursued by a massive ‘infrastructure arms race’ and a shift from models that simply talk to models that reason and act.
The competition to be superior comes at the cost of collecting users’ private data to train AI models. Key players, including Google, Meta (which owns Facebook, Instagram, Threads and WhatsApp), and Microsoft’s LinkedIn, are using their platforms to collect private data from users’ profiles or emails.
The ability for users to control or disable AI features has become a significant point of differentiation among tech platforms. While Google and LinkedIn do provide an option to opt out of their AI features, Meta does not allow this convenience.
Let us understand how different companies have been harvesting users' data:
1. Google
Google’s AI platform Gemini collects data when users engage in searches and prompts across Gemini apps, including its mobile app, Gemini in Chrome and Gemini in other web browsers. This includes any videos or photos users upload to Gemini.
Google also harvests data through interactions with apps such as YouTube and Spotify, provided users grant permission. Similarly, data from messaging and phone call apps, including call logs and message logs, can be collected with user consent.
However, on November 5, Google elevated its data harvesting practices by announcing that Gemini Deep Research can now connect to users’ other Google products, including Gmail, Drive, and Chat. This means Gemini can access Gmail messages as well as data from Docs, Drive and Chat.
Google argues that users can create more comprehensive reports by pulling information directly from their Gmail, Drive (including Docs, Slides, Sheets and PDFs), and Google Chat, alongside various sources from the web.
Does Google allow an opt-out option for data harvesting?
The tech does provide options to opt out of their data harvesting for training of AI models.
A smarter way for the users could be to use Gemini without signing in or opt for ‘temporary’ chats. This approach prevents Gemini from saving chat history.
Another option is to turn off Gmail’s smart features in your settings. This stops Google’s AI from accessing your Gmail content. However, it’s important to note that this only affects Gmail access and does not prevent Google’s access through the Gemini app itself.
For broader protection, you can turn off smart features in your Google Workspace settings. This disables AI access across Gmail, Drive and Meet.
It is important to note that following an October 2024 policy change, users must now manually adjust their privacy settings to disable Gemini’s access to private content in Gmail, Chat and Meet. Also, when Gemini prompts users for permission to access specific data, such as emails, you can simply deny these permission requests through the pop-up notifications.
However, there are important limitations to be aware of.
Turning off Gmail’s smart features does not stop Google’s access through the Gemini app, meaning users must separately control settings across different Google products. Even with these opt-outs in place, one must remain vigilant and actively deny permissions whenever Gemini requests access to specific data sources.
2. Meta
Meta, the largest owner of multiple social media platforms, has long maintained that it does not breach users’ trust by harvesting encrypted conversations. The company has consistently stated that it does not access private messages on Instagram, WhatsApp or Messenger to train its AI systems.
However, in a significant policy change announced to take effect on December 16, Meta revealed it will begin analysing direct messages, including conversations, photos and voice messages fed into its AI Chatbot.
This data will be used to customise users’ content and advertisements based on their interactions with Meta AI. For instance, if a user enquires about a ‘3-day trip to Bali’ with Meta’s AI chatbot, it could trigger targeted ads for flights, luggage or hotels across its platforms.
Meta acknowledges that AI conversations may involve sensitive topics such as religious views, sexual orientation, and racial or ethnic origin. The company claims its system is designed to prevent these interactions from being used for advertising purposes.
Additionally, Meta states that when users engage with Meta AI using voice features, the microphone will only be activated with explicit user permission.
Despite these assurances regarding private messages, Meta is actively collecting user data from public content (posts in public mode) to train its AI models. This includes photos, posts, comments, and reels that users have set to ‘public’ mode across all Meta platforms.
The scope of Meta’s data collection extends beyond its own users.
The tech company has disclosed that its AI may use information about people who do not even have Meta accounts if their information appears in other users' public posts. For example, if a Meta user mentions a non-user in a public image caption, that photo and caption could be incorporated into Meta AI’s training data.
Does Meta allow an opt-out option for data harvesting?
No, there is no way to prevent data from public posts or conversations using the chatbot from being used by Meta AI.
Meta also does not provide a facility to deactivate Meta AI in Instagram, Facebook or Threads. WhatsApp users can deactivate the option to talk with Meta AI in their chats, but this option is available only per chat, meaning that you must deactivate the option in each chat’s advanced privacy settings.
3. LinkedIn
Microsoft’s LinkedIn has not issued any notification specifically regarding India.
However, the company announced that starting from November 3, it will use certain user data to train its AI models designed to enhance user experience and better connect members to opportunities in the EU, EEA, Switzerland, Canada and Hong Kong.
Despite the absence of an official announcement for Indian users, when we examined the settings of an India-based user account, we discovered that the option allowing LinkedIn and its affiliates to use personal data and user-generated content to train generative AI models was automatically enabled by default.
The platform has explicitly stated that it will utilise data such as profile details and public posts on LinkedIn for AI training purposes. However, LinkedIn has clarified that it will not harvest private messages for this purpose.
Does the platform allow an opt-out option?
Yes, it does provide an option in settings to switch off the use of users’ data for training content creation AI models.
