Meta Removed Over 50 Paid Ads Featuring AI-Generated Sexual Images of Children

01Meta下架逾50条含AI生成儿童性虐待图像的付费广告，部分曾触达数千账户

Meta removed more than 50 paid advertisements containing AI-generated child sexual abuse material or sexualized images of minors after WIRED contacted the company about findings from the Tech Transparency Project, an independent watchdog. Some ads directed users to “nudify” or “undressing” apps, which generate fake nude images of people.

TTP found the ads while investigating promotions for those apps in Meta’s ad library, a transparency tool that records advertising shown across its platforms. The library showed that the image and video ads had run on Facebook, Instagram, Messenger and Threads between November last year and early August, targeting users in the United States, the United Kingdom and more than a dozen European countries.

Their presence exposed a gap in Meta’s review process. The company says every advertisement is checked before publication, primarily by automated tools, and its rules prohibit child sexual exploitation material, nudity and sexually suggestive imagery. Yet the ads were approved and ran for periods often lasting several days. Although most were inactive by the time researchers discovered them, they remained visible in the ad library, apparently undetected, for months.

The available data does not establish their total audience. Meta said most had minimal reach, and its records showed many reached only a handful of accounts. At least one, however, reached 2,563 accounts across eight European countries. Because the library does not disclose performance in the US or every targeted country, that figure cannot be treated as the campaign’s global reach.

TTP reported its findings to the National Center for Missing and Exploited Children’s CyberTipline, which accepts reports of online child sexual abuse material. The center declined to comment on reports it receives.

After WIRED approached Meta, the company removed the advertisements for violating policies covering child sexual exploitation, abuse, nudity and adult sexual activity. Meta said many had already been disabled and that several predated recently launched AI technology intended to detect prohibited ads when they are uploaded.

The removals did not immediately close the gap. TTP initially identified about two dozen ads, then found roughly 30 more shortly before publication. Many of the later group had been published after WIRED first questioned Meta, and several were still being shown when researchers found them. The source does not establish whether each ad underwent human review, automated review or both, leaving the precise failure in Meta’s screening system unknown.

02国际刑警组织称非洲36国记录的网络犯罪案件中55%涉及AI

Artificial intelligence was involved in 55% of the cybercrime cases recorded by 36 African countries covered in INTERPOL’s African Cyberthreat Assessment Report 2026. The report does not provide each country’s case total or explain a uniform method for deciding when a case “involves AI.”

The finding comes as reliance on digital services expands across Africa, which had more than 1.1 billion mobile subscribers in 2025. Online scams were the continent’s most commonly recorded form of cybercrime that year, and INTERPOL describes the wider fraud industry as increasingly organized and cross-border.

AI is being used at several stages of that industry. Criminals deploy deepfakes and other generated content in digital sextortion and harassment campaigns. They also use AI to write convincing messages for business email compromise, or BEC, in which an attacker impersonates a trusted contact to deceive a company or individual.

Other schemes combine genuine personal data with fabricated details to create synthetic identities. According to the report, criminals have used these identities to open bank accounts, obtain mobile loans and register SIM cards while evading some biometric checks.

Recorded cybercrime-related financial losses rose from $192 million in 2024 to $484 million. Investigators attributed that increase to several factors—not AI alone—including AI-powered fraud, stolen login credentials and sophisticated social-engineering attacks designed to manipulate victims. The assessment also found that 72% of surveyed countries had identified scam centers operating within their borders, with the greatest concentrations in West and Southern Africa.

INTERPOL said the response remains constrained by weak real-time coordination among banks, telecommunications companies and law-enforcement agencies. That delay allows criminals to transfer stolen money and exploit gaps between jurisdictions before authorities can act. Many African law-enforcement bodies also remain insufficiently prepared for AI-enabled threats, the organization said.

Governments and international investigators have begun strengthening their response. Seventeen African countries introduced or updated cybercrime laws in 2025, while Senegal launched an online platform for reporting offenses involving children. Four joint operations led to more than 1,500 arrests, the seizure of hundreds of electronic devices and the recovery of over $100 million linked to cybercrime, according to Africanews.

03Hassabis卸下Google DeepMind日常管理，Kavukcuoglu接掌Gemini与前沿研究

Google is handing day-to-day control of Google DeepMind, its central AI research and development organization, to longtime executive Koray Kavukcuoglu. Google and Alphabet CEO Sundar Pichai announced that Demis Hassabis, who previously led DeepMind while also running AI drug-development company Isomorphic Labs, will become chair of Google DeepMind and chief scientist of Alphabet.

The change separates operational responsibility for Google’s AI models and products from Hassabis’s broader work on artificial general intelligence, or AGI—AI intended to perform a wide range of intellectual tasks. Kavukcuoglu will become senior vice president of Google DeepMind, report directly to Pichai and retain his position as Google’s chief AI architect.

Kavukcuoglu will oversee Gemini model development, frontier AI research, and the teams responsible for the Gemini app and its developer offerings. He is currently Google DeepMind’s chief technology officer and has spent 13 years at the organization, where Google says he established its deep-learning team and helped lead work including WaveNet and DQN.

Hassabis said he is relinquishing his daily operational duties to concentrate on the “big picture.” In his new roles, he will work with Pichai on strategic and global AGI matters while continuing to advise Kavukcuoglu and other DeepMind leaders on models and research. He will also devote more attention to Isomorphic Labs, which applies AI to drug discovery and which he will continue to lead.

The handover comes as Google reports substantial adoption of its AI products. According to Pichai, the Gemini app has surpassed 950 million monthly users, while the company’s Gemma family of openly available models has exceeded 900 million downloads. He also said Gemini models are seeing strong demand from developers and businesses.

Hassabis said new models, including Gemini 4, are progressing under Kavukcuoglu and the existing leads. Google did not disclose Gemini 4’s capabilities or release date, when the leadership transition will be fully completed, or whether reporting lines for other DeepMind executives will change.