Meta

Meta suspends AI use in Brazil amid privacy concerns

Meta has suspended the use of its generative AI (GenAI) tools in Brazil after the country’s data protection authority issued a preliminary ban on its new privacy policy. The suspension follows a decision by Brazil’s National Data Protection Authority (ANPD) to halt Meta’s policy, citing risks to users’ fundamental data rights.

ANPD’s decision arose from concerns over Meta’s use of personal data to train its AI systems without users’ explicit consent. The agency warned of ‘serious and irreparable damage’ to the rights of data subjects and imposed a daily fine of 50,000 reais for non-compliance. Meta expressed disappointment, stating that the decision is a setback for innovation and AI development in Brazil.

The controversy in Brazil reflects broader global challenges for tech companies navigating stringent data privacy laws. In regions like the European Union, similar regulatory hurdles have forced Meta and other tech giants to pause their AI tool rollouts. Human Rights Watch highlighted risks associated with personal data in AI training, noting how personal photos, including those of Brazilian children, have been misused in image datasets, raising significant privacy and ethical concerns.

Meta’s response aligns with its recent actions in Europe, where it withheld its AI models due to regulatory uncertainties. This situation underscores the tension between advancing AI technologies and adhering to evolving data protection regulations.

Meta’s AI models excluded from EU market

Meta will withhold its future multimodal AI models from customers in the EU due to a lack of clear regulatory guidance. This decision reflects a growing tension between US tech giants and EU regulators.

Meta plans to release its multimodal Llama model in the coming months, integrating video, audio, images, and text. However, these models will not be available in the EU, impacting both European companies and those offering products in the region.

The company’s larger, text-only Llama 3 model will be available in the EU. Meta’s concerns stem from compliance with the General Data Protection Regulation (GDPR), despite briefings with EU regulators and attempts to address their feedback.

The UK, with data protection laws similar to the EU, will receive the new model without regulatory delays. Meta argues that delays in Europe harm consumers and competitiveness, pointing out that other tech companies already use European data to train their models.

Meta will remove content in which ‘Zionist’ is used as a proxy term for antisemitism

Meta announced on Tuesday that it will begin removing more posts that target ‘Zionists’ when the term is used to refer to Jewish people and Israelis, rather than supporters of the political movement. This decision is based on the claim that the world can take on new meanings and become a proxy term for nationality. Meta categorises numerous ‘protected characteristics,’ including nationality, race, and religion.

Previously, Meta’s approach has treated the word ‘Zionist’ as a proxy for Jewish or Israeli people in two specific cases: when Zionists are compared to rats, reflecting antisemitic imagery, and when context clearly indicates that the word means ‘Jew’ or ‘Israeli.’ Now, Meta will remove content attacking ‘Zionists’ when it is not explicitly about the political movement and when it uses certain antisemitic stereotypes, dehumanises, denies the existence of, or threatens or calls for harm or intimidation of ‘Jews’ or ‘Israelis.’

The policy change has been praised by the World Jewish Congress. Its president, Ronald S. Lauder, stated, ‘By recognizing and addressing the misuse of the term ‘Zionist,’ Meta is taking a bold stand against those who seek to mask their hatred of Jews.’ Meta has previously reported significant decreases in hate speech on its platforms.

A recurring question during consultations was how to handle comparisons of Zionists to criminals. Meta does not allow content that compares “protected characteristics” to criminals, but currently believes such comparisons can be used as shorthand for comments on larger military actions. The issue has been referred to an oversight board. Meta consulted with 145 stakeholders from civil society and academia across various global regions for this policy update.

Meta partners with Vodafone to optimise video delivery across Europe

Meta, the owner of Instagram, has collaborated with Vodafone to optimise short-form video delivery on 11 of its European mobile networks. That effort aims to free up network capacity without compromising the viewing experience, addressing the increasing data usage driven by platforms like Instagram, TikTok, and YouTube. According to Ericsson’s 2024 Mobility Report, global mobile traffic grew by 25% in the past year, with video accounting for 73% of all mobile traffic by the end of 2023.

Telecom operators, including Vodafone, have expressed concerns about the financial burden of upgrading networks while Big Tech companies benefit from the increased data usage. A recent EU initiative to have Big Tech contribute to 5G investment has stalled. Vodafone sees its partnership with Meta as a practical solution to address the issue of fair contribution, with Meta’s video optimisation efforts leading to more efficient network resource usage.

During an April trial on Vodafone’s British network, the optimisation resulted in a low double-digit reduction in Meta’s data traffic, freeing up capacity in busy areas such as shopping centres and transport hubs. Both Vodafone and Meta plan to continue collaborating to enhance network efficiencies further. Similarly, Spain’s Telefonica has also started working with Meta to optimise video traffic delivery on its networks.

WhatsApp introduces Meta AI for avatar creation

WhatsApp is developing a new AI feature to create user avatars, following in the footsteps of Meta AI. According to WABetaInfo, this ‘Imagine Me’ feature will allow users to generate AI-based avatars by typing prompts in their chats. The feature was discovered in the latest WhatsApp beta for Android 2.24.14.13, available through the Google Play Beta programme.

Users can generate avatars by typing ‘Imagine me’ in the Meta AI chat or ‘@Meta AI imagine me’ in other chats. A screenshot from WABetaInfo shows how this feature might look. Once enabled, users must take setup photos, and the AI will create images based on the provided prompts. The resulting images are automatically shared in the conversation, with user privacy preserved.

The feature is optional and requires users to opt in through their settings. While currently available only in limited countries, it is still under development and cannot yet be tested by all users. WhatsApp aims to make Meta AI a more integral part of daily user interactions with this innovative avatar creation tool.

Meta’s Threads hits 175 million users in the first year

According to CEO Mark Zuckerberg, Meta Platforms’ latest social media app, Threads, has reached over 175 million monthly active users just before its first anniversary. Launched on 5 July last year, Threads aimed to attract users from Twitter, now rebranded as X, during its tumultuous acquisition by Elon Musk. The app quickly gained 100 million users within a week, partly due to its integration with Instagram, but some early users eventually left.

Despite its rapid user growth, Threads has struggled with engagement. Market intelligence firm Sensor Tower reports a significant decline in user activity, with the average sessions and time spent on the app dropping considerably since its launch. Threads has yet to introduce advertising, making little to no revenue for Meta. The platform’s recent integration into the Fediverse, which allows interaction across various social media sites, has yet to boost engagement substantially.

Analysts point out that Threads needs a more precise identity and original content, which could hinder its growth. There is ongoing speculation about whether Meta will maintain Threads as a standalone app or integrate it further with Instagram. Despite these challenges, advertisers’ interest in Threads remains high.

Meta’s new strategy: AI-powered gaming experiences

Meta is set to integrate more generative AI technology into its virtual, augmented, and mixed-reality games, aiming to boost its struggling metaverse strategy. According to a recent job listing, the company plans to create new gaming experiences that change with each playthrough and follow unpredictable paths. The initiative will initially focus on Horizon, Meta’s suite of metaverse games and applications, but could extend to other platforms like smartphones and PCs.

These developments are part of Meta’s broader effort to enhance its metaverse offerings and address the financial challenges faced by Reality Labs, the division responsible for its metaverse projects. Despite selling millions of Quest headsets, Meta has struggled to attract users to its Horizon platform and mitigate substantial operating losses. Recently, the company began allowing third-party manufacturers to license Quest software features and increased investment in metaverse gaming, spurred by CEO Mark Zuckerberg’s growing interest in the field.

Meta’s interest in generative AI is not new. In 2022, Zuckerberg demonstrated a prototype called Builder Bot, which allows users to create virtual worlds with simple prompts. Additionally, Meta’s CTO, Andrew Bosworth, has highlighted the potential of generative AI tools to democratise content creation within the metaverse, likening their impact to that of Instagram on personal content creation.

Generative AI is already making waves in game development, with companies like Disney-backed Inworld using the technology to enhance game dialogues and narratives. While some game creators are concerned about the impact on their jobs, Meta is committed to significant investments in generative AI, even though CEO Zuckerberg cautioned that it might take years for these investments to become profitable.

Brazil halts Meta’s new privacy policy for AI training, citing serious privacy risks

Brazil’s National Data Protection Authority (ANPD) has taken immediate action to halt the implementation of Meta’s new privacy policy concerning the use of personal data to train generative AI systems within the country.

The ANPD’s precautionary measure, announced in Brazil’s official gazette, suspends the processing of personal data across all Meta products, extending to individuals who are not users of the tech company’s platforms. The regulatory body, operating under Brazil’s Justice Ministry, has imposed a daily fine of 50,000 reais ($8,836.58) for any directive violations.

The decision by the ANPD was motivated by the perceived ‘imminent risk of serious and irreparable or difficult-to-repair damage to the fundamental rights of affected individuals.’ As a result, Meta is mandated to revise its privacy policy to eliminate the segment related to the processing of personal data for generative AI training. Additionally, Meta must issue an official statement confirming the suspension of personal data processing for this purpose.

In response to the ANPD’s ruling, Meta expressed disappointment, characterising the move as a setback for innovation and predicting a delay in delivering AI benefits to the Brazilian population. Meta defended its practices by pointing to its transparency policy compared to other industry players who have used public content for training models and products. The company asserted that its approach aligns with Brazil’s privacy laws and regulations.

Meta lifts ban on ‘Shaheed’ after review

Meta Platforms, the parent company of Facebook and Instagram, announced it would lift its blanket ban on the word ‘shaheed’ (which means ‘martyr’ in English). The move follows a year-long review by Meta’s independent oversight board, which concluded that the company’s approach to the word was overly broad.

Meta faced strong criticism for its content policies, particularly regarding the Middle East. A 2021 study commissioned by Meta Platforms highlighted the adverse human rights impact on Palestinians and other Arabic-speaking users. Criticisms intensified with the escalation of hostilities between Israel and Hamas in October.

Why does it matter?

The review revealed that Meta’s policy on the word ‘shaheed’ did not consider its various meanings, often resulting in the removal of non-violent content. Meta acknowledged these findings and adjusted its approach, removing content only when ‘shaheed’ is paired with otherwise violating content. The oversight board welcomed this change, noting that the previous policy had led to widespread censorship across Meta’s platforms.

Meta responds to photo tagging issues with new AI labels

Meta has announced a significant update regarding using AI labels across its platforms, replacing the ‘Made with AI’ tag with ‘AI info’. This change comes after widespread complaints about the incorrect tagging of photos. For instance, a historical photograph captured on film four decades ago was mistakenly labelled AI-generated when uploaded with basic editing tools like Adobe’s cropping feature.

Kate McLaughlin, a spokesperson for Meta, emphasised that the company is continuously refining its AI products and collaborating closely with industry partners on AI labelling standards. The new ‘AI info’ label aims to clarify that content may have been modified with AI tools rather than solely created by AI.

The issue primarily stems from how metadata tools like Adobe Photoshop apply information to images, which platforms interpret. Following the expansion of its AI content labelling policies, daily photos shared on Meta’s platforms, such as Instagram and Facebook, were erroneously tagged as ‘Made with AI’.

Initially, the updated labelling will roll out on mobile apps before extending to web platforms. Clicking on the ‘AI info’ tag will display a message similar to the previous label, explaining why it was applied and acknowledging the use of AI-powered editing tools like Generative Fill. Despite advancements in metadata tagging technology like C2PA, distinguishing between AI-generated and authentic images remains a work in progress.