日本語
2026-10-29 Evening edition
Evening editionAI News Daily (Recap Edition)
2026-10-29
Only one confirmed new item was found today, so under policy this is a recap edition covering that single new item plus two Anthropic research updates.
2026-10-29 Evening edition
About this recap edition
In the 2026-10-29 06:00-18:00 JST collection window, only one confirmed new item was found (setting aside a Renesas Q3 earnings article on domestic AI data-center demand, which is not used as a calendar event). Because this meets the "one or fewer confirmed new items -> recap edition" rule, today ships as a recap edition.
- Confirmed new item: 1 OpenAI "dot" feature update
- Added recap items: 2 Anthropic research announcements
2026-10-29 Evening edition
Today's Highlights
Feature UpdateOpenAI brings text control of ChatGPT's "dot" to iMessage/RCS
2026-10-29
AI-Discovered Vulnerabilities (Recap)Anthropic: unreleased research Claude model builds attack chains against major browsers and OSes on its own
2026-10-27 09:00
Research (Recap)Anthropic expands "AI for Science" program to 20 foundations and research institutions
2026-10-28 09:00
2026-10-29 Evening edition
OpenAI brings text control of ChatGPT's "dot" to iMessage/RCS
Published: 2026-10-29
- Launched a beta for US ChatGPT Pro users letting them text the "dot" agent feature over iMessage or RCS
- Dot keeps the same memory and plugin context it uses inside the app
- Users can request tasks and check progress straight from their regular messaging app
Why it matters: This moves AI agents closer to living inside everyday chat apps, lowering the barrier to weaving AI into both work communication and personal task management.
2026-10-29 Evening edition
(Recap) Anthropic: unreleased research Claude model builds attack chains against major browsers and OSes on its own
Published: 2026-10-27 09:00 (Recap)
- Anthropic's frontier red team reports an unreleased research Claude model autonomously built five attack chains targeting Chrome, Safari, Android, Windows, and Linux
- All testing happened in lab conditions, not production; discovered flaws were reported to and fixed by vendors beforehand
- Some techniques reached outcomes like remote code execution with little to no user interaction
Why it matters: Anthropic itself demonstrated that an AI model can assemble attack chains spanning multiple vendors, pressing defenders to rethink patch speed and evaluation practices.
KEY CONTEXTTargets covered five platforms: Chrome, Safari, Android, Windows, and Linux. All testing was lab-only, and vulnerabilities were reported to and fixed by vendors before disclosure.
2026-10-29 Evening edition
(Recap) Anthropic expands "AI for Science" program to 20 foundations and research institutions
Published: 2026-10-28 09:00 (Recap)
- Expanded the "AI for Science" program, which gives researchers free access to Claude, adding 20 new funders, institutions, and research projects including NSF, the Simons Foundation, and EMBL
- Says it has supported hundreds of research projects since the program launched in May 2025
Why it matters: The AI company is shifting focus from individual researcher support to organization- and foundation-level partnerships, marking a new stage in AI's reach across scientific research.
2026-10-29 Evening edition
Trend Overview
- Major AI companies are releasing a string of studies showing AI models discovering and exploiting vulnerabilities themselves, making defenders' response speed a growing talking point
- Real-world semiconductor and data-center demand is now showing up in the earnings of Japanese companies too
- AI agents are being woven into everyday messaging channels (iMessage/RCS and the like), advancing their shift toward constant presence
2026-10-29 Evening edition
Summary
3 Key Takeaways
- Only one confirmed new item today, so this ships as a recap edition
- AI agents continue moving into everyday messaging apps
- AI-driven vulnerability discovery research is becoming a defense-readiness talking point
What to Watch Next
- How far OpenAI rolls out "dot" on iMessage/RCS
- How vendors respond to Anthropic's vulnerability-discovery research
- Concrete research outcomes from the expanded "AI for Science" program