日本語
Sources: [1] [2] [3] [4] [5] [6]
2026-09-19 Evening edition

Evening edition / 2026-09-19

AI News Daily 2026-09-19 / Review

This edition revisits six significant developments from the current half-year, rather than presenting them as new within today's collection window. Original publication dates are retained.

Evening edition

At a glance

01 / OpenAI establishes an ongoing framework for reporting model misalignment2026-09-16
02 / OpenAI details Astra's cyber capability assessment and safeguards2026-09-01
03 / Gemini 3.8 Live combines conversation with background work2026-09-15
04 / Mistral raises €3 billion to expand research and compute2026-09-08
05 / Anthropic reports Claude's formalization of Fermat's Last Theorem2026-09-04
06 / OpenAI previews custom GPT retirement and migration to plugins2026-09-11

NEWS 01 / Evening edition

OpenAI establishes an ongoing framework for reporting model misalignment

Published: 2026-09-16

OpenAI published a framework for tracking, investigating and disclosing unexpected model behavior, alongside six reports from the previous six months. Disclosure may precede a completed investigation or fix. Individual cases do not establish how frequently such behavior occurs across its models.

Disclosure practices and unresolved issues can inform vendor evaluation alongside performance. Case counts should not be mistaken for incident rates.
Tier 1・Official source — OpenAI

NEWS 02 / Evening edition

OpenAI details Astra's cyber capability assessment and safeguards

Published: 2026-09-01

OpenAI assessed Astra at the Critical cyber capability level under its Preparedness Framework and described pre-release safeguards. Monitoring covers both misuse and unauthorized model actions. The company acknowledged that false positives can delay or stop legitimate work.

Safety monitoring can explain some agent interruptions. Capability assessments should be distinguished from normal deployment settings, with clear authorization boundaries in use.
Tier 1・Official source — OpenAI

NEWS 03 / Evening edition

Gemini 3.8 Live combines conversation with background work

Published: 2026-09-15

Google introduced Gemini 3.8 Live and Live Extended Thinking for more complex tasks. They can run tools and APIs while conversation continues; Extended Thinking reasons and speaks in parallel. The article was first published September 15 and updated September 17. This is a retrospective item, not today's release.

Voice AI is expanding into task execution during conversations. Choosing which actions to delegate becomes part of both experience and operational design.
Tier 1・Official source — Google

NEWS 04 / Evening edition

€3B

Mistral raises €3 billion to expand research and compute

Published: 2026-09-08

Mistral announced a €3 billion Series D at a post-money valuation above €21 billion. Samsung Electronics led the round. Funding will support research, training compute, infrastructure and international growth, alongside its strategy for organizational control of data, models and computing environments.

AI selection also involves data control and infrastructure choices. The funding indicates the scale of investment behind that strategy.
Tier 1・Official source — Mistral

NEWS 05 / Evening edition

Anthropic reports Claude's formalization of Fermat's Last Theorem

Published: 2026-09-04

Anthropic reported that Claude formalized a proof of Fermat's Last Theorem in Lean over 11 days, producing a computer-checkable result. Multiple agents worked from established mathematics. This was not a claim to have first solved a previously unproved theorem.

The significant development is automation of verifiable reasoning. Persuasive explanations and formally checked correctness remain different standards.
Tier 1・Official source — Anthropic

NEWS 06 / Evening edition

OpenAI previews custom GPT retirement and migration to plugins

Published: 2026-09-11

OpenAI outlined plans to retire custom GPTs and provide a migration path to plugins combining reusable instructions with connected apps. Existing GPTs remain usable until their applicable retirement date. Timing and availability differ by plan and workspace, so users must check their own notices.

An inventory of GPTs, creators and users supports migration preparation. The announcement is a plan, not confirmation that all GPTs have already retired.
Tier 1・Official source — OpenAI release notes

Evening edition

What connects these stories

This edition revisits six significant developments from the current half-year, rather than presenting them as new within today's collection window. Original publication dates are retained.

Voice-driven work and proof formalization are advancing alongside disclosure and monitoring of model misalignment.

Model selection should consider verification, data control and migration plans as well as performance.