Meta authorizes use of public Instagram profile photos to train AI

What happened?

Meta, the parent company of Facebook, Instagram, and WhatsApp, has updated its privacy policy to explicitly authorize the use of profile photos and public posts from Instagram in training its artificial intelligence (AI) models. The move, initially reported by BBC Technology on June 4, 2024, allows the company to use publicly shared images on the platform to improve its generative AI algorithms, including text-to-image models that compete with OpenAI's DALL-E or Google's Imagen.

According to Meta's official statement, users can opt out through privacy settings, but this option is not available for business or creator accounts. The company claims it will only use public content, excluding private messages and posts from private accounts. However, the definition of 'public' includes profile photos, posts with public visibility, public stories, and hashtags. Meta has also stated it will not use content from users under 18, though it is unclear how it will effectively verify age.

Why is this important?

This decision marks a milestone in the relationship between big tech companies and user privacy. Unlike other companies like OpenAI, which train their models with data obtained from public sources like the web (e.g., Common Crawl) or through licensing agreements with Shutterstock, Meta is directly using user-generated content from its own platforms. This raises fundamental questions about consent and data ownership, especially since users do not have granular control over which specific images are used.

For Instagram users, millions of profile photos and public posts could be used without direct compensation. Privacy activists have called the measure a 'recipe for disaster,' according to a statement from a Big Brother Watch spokesperson to the BBC. Additionally, organizations like the Electronic Frontier Foundation (EFF) have pointed out that the opt-out option is insufficient, as it requires users to know about its existence and navigate complex menus, and is not available for business accounts. This move also contrasts with the approach of other platforms: for example, Reddit and Twitter have begun charging AI companies for access to their data via APIs, while Meta opts for a free but controversial access model.

Consequences for businesses and users

For businesses that rely on Instagram for marketing, this means their brand images could be used to train AI models that compete with their own products or services. For example, a fashion brand that posts photos of its collections on Instagram could see those images integrated into a generative AI model that others then use to create similar designs without attribution. Individual users, especially those with public profiles, will have their personal data integrated into AI systems without direct control over its future use, which could lead to deepfakes or unauthorized uses of their image.

Meta has tried to mitigate criticism by offering the opt-out option, but the process is not straightforward: users must go to Settings > Privacy > Data Options and disable 'Use of data for AI.' Additionally, the company has not clarified whether data already used before opting out will be removed from training sets. For business and creator accounts, the option simply does not exist, affecting millions of small businesses and public figures who depend on Instagram for their activity.

'It's a recipe for disaster,' said a Big Brother Watch spokesperson. 'Meta is using people's photos without their explicit consent, and the opt-out option is insufficient.'

From a regulatory standpoint, this policy could violate the European Union's General Data Protection Regulation (GDPR), which requires a clear legal basis for processing personal data. Meta has delayed implementation in the EU and UK due to these concerns, but in other regions like the United States and Latin America, the policy is already active. The U.S. Federal Trade Commission (FTC) and the European Commission have announced investigations into similar practices by other companies, so Meta could face significant fines if found to have violated regulations.

What should readers know?

If you have a public Instagram account, your profile photo and other public images could be used to train AI. To opt out, go to Settings > Privacy > Data Options and disable 'Use of data for AI.' However, this option is not available for business or creator accounts. Meta claims it will only use public content, but the definition of 'public' includes profile photos, posts with public visibility, and hashtags. Private accounts are not affected.

This policy takes effect immediately in most regions, although in the European Union and the UK, Meta has delayed implementation due to stricter privacy regulations. Users in those regions have additional protection under GDPR, but should still be vigilant about future changes.

Additionally, it is important to note that Meta has not specified whether data used to train models will be anonymized or whether it could be extracted from the final model through inversion techniques. Recent research has shown that it is possible to reconstruct training images from generative models, posing additional privacy risks.

Long-term implications

This move by Meta could accelerate regulation on the use of personal data in AI. The FTC and European Commission are already investigating similar practices, and new laws requiring explicit consent for the use of personal data in AI training are likely to emerge. Additionally, it could incentivize users to migrate to more privacy-respecting platforms like Signal, Telegram, or decentralized platforms like Mastodon. However, Instagram's network effect makes mass migration difficult.

In the competitive landscape, Meta gains a significant advantage by having access to a massive set of labeled visual data (over 2 billion monthly active users on Instagram), which could improve its generative AI models compared to competitors like Google and OpenAI, which rely on public or licensed web data. This could translate into more powerful AI products, such as more realistic image generation or virtual assistants with better visual understanding. However, the reputational cost could be high, especially if privacy scandals similar to Cambridge Analytica occur.

In summary, Meta's decision to use Instagram photos to train AI represents a turning point at the intersection of social media and artificial intelligence. Users should be aware of their rights and take steps to protect their privacy, while regulators and society at large must debate the ethical limits of this practice. Time will tell whether this move drives greater regulation or sets a precedent that other companies follow.