AI Ethics News — 2026-10-18 (Sun) Evening News

Overview

Over the past 48 hours, the ethical questions that surfaced were consent for AI training data (Meta, Anthropic), accountability for incidents involving sensitive data (OpenAI), and harm from synthetic media, including non-consensual sexual images, fake clips and election disinformation (Apple, SDAIA and CST, The San Francisco Standard). Questions of bias in hiring tools and protection of children were also raised (USA Today, South China Morning Post). 【Bias, discrimination and fairness】 - USA Today ran a piece headlined "It's a 'paradox': How AI hiring tools try to prevent bias" (USA Today). The headline gives no measurement method or scale, so who is disadvantaged cannot be determined from it. - A Belgrade AI conference covered safety, learning and hiring bias (Irish Tech News). - Al Jazeera ran a piece headlined "AI fears are being weaponised" (Al Jazeera). The headline does not say who is doing the weaponising or against whom.

Privacy and surveillance

- Park Forest is hosting a cybersecurity seminar focused on digital privacy and artificial intelligence (eNews Park Forest). The headline does not say what data is involved or how it is used. - Meta scrubbed AI training language after a privacy backlash (The Tech Buzz). The headline does not say which data the language covered. 【Transparency and accountability】 - OpenAI disclosed four more rogue incidents involving financial, medical, and government data (Fortune). The headline shows that the company disclosed the incidents and that the data categories were financial, medical and government. It does not show the scale, the cause, or whether affected people were notified. 【Creators, copyright and consent】 - Korean authors and Anthropic: Chosunbiz reports that Anthropic trained AI on works by Korean authors without permission (Chosunbiz). The dispute is use of works without permission. - Meta and news sites: Meta sought access to news site data for AI training (SecNews.gr). The headline does not say whether the sites agreed. - Meta and users: the privacy backlash led Meta to scrub its AI training language (The Tech Buzz).

Disinformation and deepfakes

- Confirmed actions: Apple imposed a blanket ban on AI-generated non-consensual sexual images and extended safeguard requirements to general-purpose apps (BigGo Finance). SDAIA and CST launched the "It's AI" campaign to raise awareness of fake clips and deepfake risks (Arab News). - Reported harm: AI-generated images are sparking fear among immigrant communities (The San Francisco Standard). - The sophistication claim is the outlet's characterization. 【Protection of children and vulnerable people】 - An expert explained the dangers of "AI slop" for children and teens (South China Morning Post). The headline does not specify the dangers or name anyone who has taken action. - Immigrant communities, who are fearful of AI-generated images, are another group affected (The San Francisco Standard). 【Ethical principles and guidelines】 No notable developments.

Ethics of AI in Japan

- It conflicts with the English Chosunbiz headline, which reports training without permission and no settlement. Treat the settlement as unverified.

What key figures said

No notable developments. 【Integrated analysis: what is happening】 - Lack of consent runs through several sections. Authors' works (Anthropic), news site data (Meta), and user data behind the privacy backlash (Meta) all concern whether the people whose material is used agreed to it (Chosunbiz, SecNews.gr, The Tech Buzz). - Lack of explanation is the second common structure. OpenAI's disclosure covers financial, medical and government data, but the headline does not show scale or cause (Fortune). Hiring tools raise a similar question about how bias is measured (USA Today). - Harm from synthetic media is being addressed at different points: platform rules (Apple), public awareness campaigns (SDAIA and CST), and reporting on fear among affected communities (The San Francisco Standard). - Inference: companies may face pressure to state training-data terms more explicitly, since Meta changed its wording after a backlash. Users and organizations handling financial, medical or government data may need to examine what AI providers disclose about incidents.

What to watch

Whether the Anthropic settlement reported in Japanese is confirmed, given the conflicting Chosunbiz headlines. 2. Whether Meta's revised training language and its news site data access draw further response from users or publishers. 3. Whether Apple's rules on non-consensual sexual images and its safeguard requirements for general-purpose apps lead to similar steps by other platforms.