Meta Uses Your Instagram Photos to Train AI Without Your Permission

Meta, the parent company of Instagram and Facebook, has enabled by default the use of images from public Instagram accounts to train its generative artificial intelligence model, Muse Image AI. The move, reported by Engadget, allows the company to use visual content from millions of users without their explicit consent, unless they manually opt out. This action adds to a growing trend among big tech companies to leverage user data for training AI models, but with a particularly aggressive approach: opt-out instead of opt-in.

What Happened?

According to information published by Engadget, Meta has updated its terms of service to include the use of Instagram photos in training its AI models. The option is enabled by default for all public accounts, meaning users must search in privacy settings and manually disable it if they do not want their images used. Engadget notes that the setting is located in 'Settings' > 'Privacy' > 'Data Use for AI', though the exact location may vary depending on the app version. Meta has confirmed that private accounts are not affected, but any public account — including those of businesses, influencers, and regular users — is automatically included.

"Meta has automatically opted all public accounts into its Muse Image AI" — Engadget

Muse Image AI is a generative image model that Meta announced in 2023, designed to create and edit images based on textual descriptions. Unlike other models such as OpenAI's DALL-E or Midjourney, Muse is trained with proprietary Meta data, including images from Instagram and Facebook. The company claims that using these images improves the model's quality and relevance, but has not specified what proportion of the training set comes from Instagram.

Why Is This Important?

This move by Meta reignites the debate over personal data collection for artificial intelligence. Unlike other companies that have sought explicit consent, Meta opts for an opt-out model, which may result in most users being unaware that their photos are being used. According to a 2023 Pew Research Center study, 62% of social media users do not review changes in privacy policies, suggesting that a large majority may not know their images are being used. This raises serious questions about transparency and the control users have over their own content.

Furthermore, Meta's approach contrasts with that of other tech companies. For example, OpenAI requires ChatGPT Plus users to explicitly consent to the use of their data for training, and Google allows Workspace users to disable data use for improving its AI models. In contrast, Meta has opted for a more aggressive strategy, similar to what it used with facial recognition in the past, which ultimately resulted in fines and legal settlements.

Potential Consequences

• Privacy: Users may feel their content is being exploited without permission, eroding trust in the platform. This is especially concerning for influencers and professionals who rely on Instagram for their business, as their images are their main asset.
• Regulation: It could attract the attention of data protection regulators, especially in the European Union under GDPR. Recall that in 2023, the EU fined Meta 390 million euros for privacy violations related to targeted advertising. Using images for AI could be considered new data processing requiring explicit consent under GDPR.
• Competition: Other platforms like X (formerly Twitter) or TikTok could follow suit, normalizing the use of user data for training AI without consent. In fact, X already announced in 2023 that it would use public posts to train its Grok model.
• Model Quality: Using real Instagram images could improve Muse Image AI's ability to generate realistic images, but at the cost of privacy. Meta claims images are anonymized and aggregated, but critics point out that personal information can be reconstructed from metadata.
• Impact on Creators: Artists and photographers who publish their work on Instagram could see their styles replicated by AI without compensation, reigniting the debate over copyright in the era of generative AI.

What Should Readers Know?

If you have a public Instagram account and do not want your photos used for AI training, you must go to privacy settings and disable the corresponding option. The exact steps vary by app version, but it is generally found in 'Settings' > 'Privacy' > 'Data Use for AI'. However, note that even if you disable this option, Meta may have already used your previous photos. Additionally, the option may be hidden or hard to find; Engadget suggests some users have reported that the setting does not appear in all regions.

It is advisable to periodically review privacy settings, as platforms can change their policies without notice. Also consider that private accounts are not automatically included, but if you ever switch your account to public, your images could be used retroactively. For greater protection, some experts recommend not posting sensitive content on public accounts and reviewing the terms of service of the platforms you use.

Historical Context

This is not the first time Meta has faced criticism over user data use. In 2018, the Cambridge Analytica scandal revealed how Facebook data was used without consent for political purposes, leading to a $5 billion FTC fine and changes in privacy policies. Now, using images for AI poses similar risks, but with broader technological implications. Unlike Cambridge Analytica, where data was used for political targeting, here data is used to train a model that could have commercial applications, such as generating advertising content or photo editing.

Another relevant precedent is Facebook's facial recognition program, which was discontinued in 2021 after years of criticism and lawsuits. In that case, Meta also used an opt-out model and eventually agreed to pay $650 million to settle a class-action lawsuit in Illinois for violations of biometric privacy law. This precedent suggests that the new use of images for AI could face similar legal challenges.

Globally, the European Union is advancing AI regulation with the AI Act, which classifies AI systems by risk. Using personal data to train generative models could be considered high-risk, forcing Meta to conduct impact assessments and obtain explicit consent. Meanwhile, in the United States, the FTC has shown interest in regulating data use for AI, and in 2023 issued a warning to companies about using user data to train models without transparency.

In conclusion, Meta's decision to enable by default the use of Instagram images to train Muse Image AI represents another step in the growing commodification of personal data. While the company argues it improves its services, the opt-out approach and lack of transparency could have long-term legal and trust consequences. Users should stay vigilant and take steps to protect their privacy, while regulators will likely intensify scrutiny of these practices.