日本語
2026-09-11 Morning edition
Morning edition
AI News Daily 2026-09-11
Anthropic details how it detects and blocks real-world AI misuse, Oracle's cloud bookings surge on AI demand, and OpenAI and DeepSeek both reshape their plans and models.
Today's Highlights
Company News
Anthropic publishes a misuse-defense report covering seven categories of abuse, including bioweapon-related research and military drone software
Published: 2026-09-10
Company News
Oracle's cloud revenue jumps 47% as it raises full-year revenue guidance to $90 billion on AI demand
Published: 2026-09-10
Feature Update
OpenAI pauses new sign-ups and upgrades for the $200/month ChatGPT Pro plan
Published: 2026-09-10
Model Release
DeepSeek ships its first native multimodal model, DeepSeek-V4.1-Flash, via API
Published: 2026-09-10
Anthropic Discloses Seven Categories of AI Misuse It Blocked
Published: 2026-09-10
7 categories
- Covers cases detected and blocked between December 2025 and August 2026
- Five cases involved research that could be repurposed for bioweapons, including help with grant applications for gain-of-function research on avian flu
- Also disclosed: assistance to Russian autonomous military drone control software, plus cyberattacks, influence operations, fraud, conventional weapons development, and model distillation
Why it matters: This is primary-source evidence of how well frontier AI safety measures detect and block misuse in live operation, useful for companies shaping their own AI usage policies and monitoring.
Oracle Raises Full-Year Guidance to $90B on AI Demand
Published: 2026-09-10
93%
- Q1 FY2027 results (announced Sept 10, 2026): cloud revenue up 47% year over year to $9.9 billion, with cloud infrastructure (OCI) revenue up 93%
- Raised its FY2027 revenue guidance to $90 billion on the back of large AI-related contracts
- Shares jumped sharply following the earnings release
Why it matters: Demand for AI infrastructure is starting to show up in hard order and revenue numbers, giving companies a concrete data point for cloud investment and vendor decisions.
OpenAI Pauses New ChatGPT Pro Sign-Ups
Published: 2026-09-10
$200
- As of September 10, 2026, OpenAI paused new sign-ups for ChatGPT Pro (the $200/month "Pro 20X" plan)
- Upgrades from Free, Go, Plus, and the $100 Pro plan are also paused
- Existing $200 Pro subscribers, and existing or new $100 Pro subscribers, are unaffected
Why it matters: With new sign-ups to the top-tier plan restricted, enterprises and power users planning heavy usage need to hold onto existing contracts or consider switching to another plan or the API.
DeepSeek Ships Its First Native Multimodal Model
Published: 2026-09-10
552B
- DeepSeek-V4.1-Flash is DeepSeek's first native multimodal model, supporting text and image input and output by default via the API
- Uses a new 552B-parameter MoE architecture with 8B active parameters for input and 16B for output
- The prior V4-Flash and V4-Flash-Vision-Exp models are being phased out, with existing requests automatically routed to V4.1-Flash
Why it matters: Low-cost open-weight vendors are rapidly strengthening multimodal support, which could shift the cost structure and model choice for image-handling applications.
Trend Overview
- Frontier AI vendors are advancing safety transparency (Anthropic's misuse report) and fine-tuning plans and pricing (OpenAI's plan limits, DeepSeek's new pricing) at the same time
- Demand for AI infrastructure investment is starting to show up in hard earnings numbers, as seen in Oracle's surging cloud bookings
- Open-weight players like DeepSeek continue to compete on both price and features, adding multimodal support alongside flexible time-of-day pricing
Wrap-Up
Three things to remember
- Anthropic's misuse-defense report shows frontier AI safety controls operating at real-world scale, including bioweapon-related research and military drone cases
- Oracle's Q1 results (cloud revenue +47%, OCI +93%) show AI infrastructure demand is now backed by real bookings and revenue
- OpenAI paused new ChatGPT Pro sign-ups while DeepSeek shipped its first native multimodal model — plans and model lineups keep shifting fast
What to watch next
Whether other frontier labs follow Anthropic's lead on misuse transparency reporting; how Oracle's raised guidance holds up against AI-spending sustainability concerns; and whether OpenAI reopens ChatGPT Pro sign-ups as DeepSeek's new multimodal model and pricing pressure competitors.