日本語2026-09-20 Morning edition
MORNING NEWSAI News Daily
2026-09-20
AI that decides, AI that enters workflows
Today’s highlights
AI safety & policyOpenAI publishes an Australian youth AI safety blueprint
2026-09-18
Generative AI & creative workGoogle Flow enters New York Fashion Week production workflows
2026-09-18 22:00
AI safety & advertisingMeta expands AI-based scam-ad defenses in Poland
2026-09-18 23:55
AI & marketingGoogle explains how to train ad AI on closed-won outcomes
2026-09-18 21:00
Agentic AIJev becomes Vercel AI Gateway’s fastest-adopted model
2026-09-18 16:00
OpenAI publishes an Australian youth AI safety blueprint
Published: 2026-09-18
- The blueprint sets six pillars including AI literacy, age-appropriate safeguards, age assurance, crisis support, and parental controls.
- ChatGPT for Teens began rolling out to Australian users aged 13–17 in August.
Why it matters: Youth AI safety is becoming a product-design requirement, with responsibility placed on providers as well as families.
Google Flow enters New York Fashion Week production workflows
Published: 2026-09-18 22:00
- Google’s Envisioning Studio and Labs co-built specialized tools with two fashion designers.
- The tools supported set design and styling work that appeared on New York Fashion Week runways.
Why it matters: The value of generative AI is shifting from one-off images to tools embedded in domain-specific creative workflows.
Meta expands AI-based scam-ad defenses in Poland
Published: 2026-09-18 23:55
- Meta says its new system removes 50% more celebrity-impersonation ads than the previous system.
- It plans identity checks for 100% of financial-services advertisers targeting Poland and says over 88% of scam ads are removed before reports.
- These are Meta-reported figures, not independent measurements.
88%+
Why it matters: Scam prevention is becoming an operating system that combines detection, advertiser verification, and reporting support.
Google explains how to train ad AI on closed-won outcomes
Published: 2026-09-18 21:00
- Google describes centralizing first-party data with Data Manager and a lead-intent scoring pilot.
- The approach feeds closed-won outcomes back into bidding systems while accounting for long sales cycles.
Why it matters: Ad AI performance depends on the business outcome used as its training signal, rather than click volume alone.
KEY CONTEXTOutcome data is the control point
Jev becomes Vercel AI Gateway’s fastest-adopted model
Published: 2026-09-18 16:00
- Vercel says nearly 13% of paid teams used Jev within 24 hours of launch.
- Jev is a System One Model that returns typed choices, probabilities, and confidence rather than conversational prose.
- The figure measures early adoption inside Vercel; it does not prove accuracy or retention.
13%
Why it matters: Agent routing and guardrails may benefit from models optimized for fast, machine-readable decisions instead of prose.
Trend view
- From answer generation to specialized workflows and machine-readable decisions.
- Safety is expanding to age protection, identity checks, and reporting support.
- Launch velocity is a signal; accuracy and retention still require observation.
Summary
Three takeaways
- Youth safeguards move into default design
- Generative AI enters domain workflows
- Jev gains early traction with typed decisions
What to watch
- How the six pillars reach products
- Independent checks of company metrics
- Jev accuracy and sustained usage