Recent AI news · Original report: 2026-10-12 (KST)

Google SynthID Detector opens to all, AI prenatal ultrasound, Meta child safety (Oct 12, 2026)

This article was compiled with AI. This English translation preserves the original report's collection date, findings, and limitations. Please consult the linked sources for current information. Originally collected: 2026-10-12 10:12 (KST, UTC+09:00).

Translated into English: 2026-10-12 (KST). The translation date does not indicate a new source check.

At a glance

This week's three announcements show AI being used in prenatal checkups, in checking online content, and in child protection. All figures are the ones reported by the organizations themselves.

This week's announcements

We opened the official posts ourselves, up to 10:12 a.m. Korea time on October 12, 2026. Dates are the ones printed on the original posts.

Prenatal ultrasound AI: eight hours of training, then AI reads the scan

On October 6, Google described this study in an interview with one of its AI researchers and its in-house OB/GYN. It is an explanation of research, not a product launch.

• How it works: It uses a "blind sweep" ultrasound, where the operator sweeps the probe over the abdomen in a set pattern. Google says healthcare workers with no ultrasound experience can learn it in eight hours. An AI model reads the video to estimate gestational age and indicate the baby's position.
• On-device processing: The post says the analysis runs on the device, so it needs neither an electricity supply nor Wi-Fi.
• Study size: The post cites 1,000 mothers in Nairobi, Kenya, and 1,000 more in Chicago. The abstract of the linked paper (JAMA Network Open, July 2026) lists 2,043 participants overall and 385 in the primary evaluation set.
• Results: According to the abstract, for pregnancies at 16 to 36 weeks the mean error in estimating gestational age was 4.2 days for the AI and 4.5 days for the standard estimate by expert sonographers.

SynthID Detector: anyone can check whether a file was made with AI

On October 7, Google said it is opening SynthID Detector, its tool for verifying AI-generated content, to everyone. SynthID is Google's technology for embedding a watermark that people cannot perceive into AI-generated content.

• What changed: The tool first appeared last year for media professionals. From October 7 it is available to anyone globally, in English.
• What it checks: Image, video, and audio files. Besides Google's own AI, it covers content made with AI from partners OpenAI, NVIDIA, and Kakao, with Apple to follow soon.
• Scale: Google says that since SynthID launched in 2023 it has watermarked more than 180 billion images and videos and 240,000 years of audio. The verification features in Search, the Gemini app, and Chrome handle more than 1 million requests a day, according to the post.

Meta: catching ads that lead to child exploitation material

On October 7, Meta published a Newsroom post, filed under its India Newsroom, on measures added to its work against child sexual exploitation. It explained that some ads look harmless on their own while directing people to illegal material off its platforms.

• New measures: A large language model (LLM) detects seemingly harmless ads that point people toward illegal material, and Meta looks at where an ad leads, not only what it shows. It also added an AI agent that probes its own defenses for weaknesses.
• Figures: Meta says it took action on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram between January and June 2026, with over 97% found before anyone reported it. In India the figures were 5.3 million pieces and over 98%.
• Existing tools: The post says matching technology such as PhotoDNA, which finds copies of already known illegal images and videos, has been in use since 2011.

What it means and for whom (our inference)

• The ultrasound research is aimed at places with too few sonographers. For pregnant women in South Korea, who get ultrasounds at OB/GYN clinics, it probably changes nothing right away.
• SynthID Detector recognizes files that carry the watermark. A result of "no watermark" probably does not mean a person made the file.
• The Meta post appeared in its India Newsroom, so the post alone does not tell us how the new measures apply to users in South Korea.

What we could not confirm

• No hands-on use: We did not use or run any of the three.
• Full text of the ultrasound paper: The JAMA Network Open page would not open, so we read the abstract in the Europe PMC database. The abstract gives no accuracy figure for fetal position, and we could not confirm why the post's "1,000 each" differs from the abstract's counts. The post also does not say whether this has been released as a product.
• SynthID Detector language support, pricing, and accuracy: Korean-language support, pricing, and accuracy are not in the announcement. The service page would not open, so we relied on two posts, from Google and Google DeepMind.
• Meta's figures: They come from Meta's post alone, and we could not check them against outside data. The post gives no rollout date or effectiveness figures for the new AI tools.

Sources

• Google, Ask a Scientist: How are researchers using AI to help pregnant women access ultrasounds? (2026-10-06) https://blog.google/innovation-and-ai/models-and-research/google-research/blind-sweep-ultrasounds-ai/
• JAMA Network Open, Generalization of AI-Based Gestational Age Assessment Using Blind Sweep Ultrasonography (2026-07) https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2851451
• Europe PMC, abstract of the paper above (PMID 42424080) https://europepmc.org/article/MED/42424080
• Google, We're making it easier to identify AI-generated content globally (2026-10-07) https://blog.google/innovation-and-ai/models-and-research/google-deepmind/synth-id-ai-content/
• Google DeepMind, SynthID overview page https://deepmind.google/models/synthid/
• Meta, Measures We've Put in Place to Fight Child Exploitation (2026-10-07) https://about.fb.com/news/2026/10/measures-weve-put-in-place-to-fight-child-exploitation/