Meta's AI labelling ‘inconsistent', internal oversight board finds
Meta's labelling of AI generated content is 'inconsistent', the tech giant's internal oversight board said in a ruling published Tuesday.
The board is an internal body which makes content moderation decisions about Meta platforms Facebook and Instagram, which said it is concerned that despite 'the increasing prevalence of manipulated content across formats, Meta's enforcement of its manipulated media policy is inconsistent'.
'It must prioritise investing in technology to identify and label manipulated audio and video at scale in order that users are properly informed,' the ruling added.
In February 2024 Meta started labelling AI generated content through watermarks or visible markers, in a bid to help users distinguish between real and fake content on its social media platforms Facebook, Instagram and Threads.
The board ruled on one case where Meta decided not to label a likely manipulated audio clip of two Iraqi Kurdish politicians discussing rigging parliamentary elections, weeks before polls opened.
The ruling said that the company's failure to automatically apply a label to all instances of the same manipulated media was 'incoherent and unjustifiable'.
It also stressed that Meta should make labels for manipulated media available in the local language available on its platforms.
The board also said it is concerned about the company's reliance on third parties for technical assessment of likely manipulated content and added that Meta should have this expertise available internally.
The EU's AI Act – legislation that regulates AI tools according to the risk they pose to society – obliges companies that create content such as deepfakes to mark their outputs as artificially generated.
Meta has been contacted for a statement.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Euronews
5 hours ago
- Euronews
Meta's AI labelling ‘inconsistent', internal oversight board finds
Meta's labelling of AI generated content is 'inconsistent', the tech giant's internal oversight board said in a ruling published Tuesday. The board is an internal body which makes content moderation decisions about Meta platforms Facebook and Instagram, which said it is concerned that despite 'the increasing prevalence of manipulated content across formats, Meta's enforcement of its manipulated media policy is inconsistent'. 'It must prioritise investing in technology to identify and label manipulated audio and video at scale in order that users are properly informed,' the ruling added. In February 2024 Meta started labelling AI generated content through watermarks or visible markers, in a bid to help users distinguish between real and fake content on its social media platforms Facebook, Instagram and Threads. The board ruled on one case where Meta decided not to label a likely manipulated audio clip of two Iraqi Kurdish politicians discussing rigging parliamentary elections, weeks before polls opened. The ruling said that the company's failure to automatically apply a label to all instances of the same manipulated media was 'incoherent and unjustifiable'. It also stressed that Meta should make labels for manipulated media available in the local language available on its platforms. The board also said it is concerned about the company's reliance on third parties for technical assessment of likely manipulated content and added that Meta should have this expertise available internally. The EU's AI Act – legislation that regulates AI tools according to the risk they pose to society – obliges companies that create content such as deepfakes to mark their outputs as artificially generated. Meta has been contacted for a statement.
Mediapart
7 hours ago
- Mediapart
Paris Olympics, Paralympics cost up to €6 billion say auditors
The cookies and similar technologies we use on Mediapart are of different natures and allow us to pursue different purposes. Some are necessary for the functioning of the site and the mobile application (you cannot refuse them). Others are optional but help to facilitate your experience as a reader and in some way support Mediapart. You can refuse or accept them below, depending on their purpose. Do you agree that Mediapart uses cookies or similar technologies for the following purposes ? You can make your choice, for each category, by enabling or disabling the switch button. Mandatory for the operation of the site or application Subscriber login, anonymized audience measurement, sending of push notifications, tracking of failures, highlighting of our services these tools are necessary to track the activity of our services and their proper functioning. read more Here are the various cookies and similar technologies included in this category : Authentication cookies : subscriber login. : subscriber login. AT Internet : anonymized audience measurement : anonymized audience measurement Display of multimedia editorial content : videos (YouTube, Dailymotion, Vimeo, INA), social networks (Facebook, Instagram, Pinterest, Twitter), documents (Scribd, Document Cloud, Slideshare), sounds (SoundCloud, Spotify, Deezer), maps (Google Maps, Mapbox, CartoDB, uMap), infographics (Highcharts, GitHub, Datawrapper, Flourish, Infogram, ThingLink, jQuery, Google Fonts, Bootstrap), live blogs (24liveblog, CoverItLive), media integration support in Journal and Club publications (Embedly). : videos (YouTube, Dailymotion, Vimeo, INA), social networks (Facebook, Instagram, Pinterest, Twitter), documents (Scribd, Document Cloud, Slideshare), sounds (SoundCloud, Spotify, Deezer), maps (Google Maps, Mapbox, CartoDB, uMap), infographics (Highcharts, GitHub, Datawrapper, Flourish, Infogram, ThingLink, jQuery, Google Fonts, Bootstrap), live blogs (24liveblog, CoverItLive), media integration support in Journal and Club publications (Embedly). Typeform : optional questionnaires to collect readers opinions on our digital products. : optional questionnaires to collect readers opinions on our digital products. Datadog (only on the website) : technical indicators and load balancing. : technical indicators and load balancing. Selligent (only on the website) : communication with the subscriber, highlighting of services, offers and benefits. : communication with the subscriber, highlighting of services, offers and benefits. Batch (only on the app) : sending push notifications and in-app messages. : sending push notifications and in-app messages. Firebase Cloud Messaging (only on app) : required for push notifications to work on Android. : required for push notifications to work on Android. Microsoft App Center (only on app) : app update and failure tracking system. Statistics These tools allow us to collect statistics on site and mobile application traffic to understand usage, detect possible problems and optimize the ergonomics of our products. read more These are the third-party tools included in this category : AT Internet : audience measurement related to subscriber ID. : audience measurement related to subscriber ID. CrazyEggs (only on website) : customer journey analysis. : customer journey analysis. Nonli (only on website) : helps our social network team to publish our contents on social networks. : helps our social network team to publish our contents on social networks. Qiota (available only on the website) : management of the datawall system. Advertising retargeting There is no advertising on Mediapart. But we do promote our content and services on other sites and social networks. For this, we use technologies made available by some advertising companies. read more These are the third-party tools included in this category : Facebook (only on the website) : audience targeting on social networks to promote Mediapart. Content Access Management We use the Qiota service from Opper Marketing Suite to configure the activation of a datawall on certain content (available only on the website). This system, intended for non-subscribed users, requires the input of an email address to access the relevant content. By providing this information, the user consents to its collection, storage, and use for statistical purposes. In accordance with the applicable regulations, users have the right to access, rectify, and delete their data, which they can exercise by contacting dpo@ Save and close
AFP
8 hours ago
- AFP
Manipulated image of burning Tel Aviv skyline circulates after Iranian strikes
"Iran hit Israel back. They invaded Tel Aviv," reads the Thai-language caption of an image shared on Facebook on June 14, 2025. The image appears to show the city's Marganit Tower and a nearby block of flats engulfed in flames (archived here and here). It surfaced after Israel attacked Iranian nuclear and military targets on June 13, prompting retaliatory strikes from Tehran (archived here and here). The arch foes continued to trade missile fire before US President Donald Trump announced a ceasefire on June 24 to end the conflict, which has killed 610 in Iran and 28 in Israel (archived here and here). The announcement came just hours after Iran launched strikes against an American military base in Qatar, which Trump described as a "weak" retaliation for US strikes against Iranian nuclear sites (archived link). Image Screenshot of the false Facebook post captured on June 20, 2025, with a red X added by AFP The same image also circulated in similar Thai-language Facebook posts, as well as in English Facebook and X posts. The circulating image, however, does not match news footage of Iranian strikes on Tel Aviv. A reverse image search on Google led to the same image in a now-deleted Facebook post from June 14, from a user who describes himself as a Libyan programmer and designer. The image's Arabic-language caption reads: Footage from the BBC shows the area in Tel Aviv was struck by an Iranian missile, but did not set off fires as seen in the false post (archived link). Image Screenshot comparison of the circulating image (left) and the BBC footage (right), with corresponding buildings highlighted by AFP Similar footage shared by France 24 on YouTube also shows plumes of smoke rising from near the impact of the strike, and no surrounding buildings engulfed in flames (archived link). AFP has previously debunked several other false claims related to the Iran-Israel conflict.
