日本語
2026-10-16 Morning edition
Morning editionAI News Daily
2026-10-16
Anthropic strengthens safety oversight by giving outside evaluator METR staff-level access, while OpenAI upgrades ChatGPT Voice to GPT-Live-1.1 for more natural, multilingual speech.
2026-10-16 Morning edition
Today's Highlights
CorporateAnthropic grants outside evaluator METR staff-level access
2026-10-15
Feature UpdateOpenAI rolls out GPT-Live-1.1 for ChatGPT Voice
2026-10-15
2026-10-16 Morning edition
Anthropic Grants METR Staff-Level Access
Announced: 2026-10-15 (publication date)
- Gave six members of nonprofit AI evaluator METR staff-level access to its hardware, offices, and internal tools.
- Shares information with METR much as it does with internal alignment-risk staff, aside from customer data and attorney-client privileged material.
- METR is designated a "Tier 3" evaluator, meaning it can publish its findings without Anthropic's prior approval.
Why it matters: Granting outside evaluators this depth of access is a test case for independently verifying AI safety claims, directly strengthening the credibility of the safety information companies rely on when adopting AI.
KEY CONTEXTMETR is a "Tier 3" evaluator, able to publish its findings without pre-approval.
2026-10-16 Morning edition
OpenAI Rolls Out GPT-Live-1.1 for ChatGPT Voice
Announced: 2026-10-15 (publication date)
- Updated ChatGPT Voice's underlying model to GPT-Live-1.1 / GPT-Live-1.1 Quip, producing more natural accents across languages.
- Added new voices for French, German, Hindi, Japanese, Korean, and Mexican Spanish, plus a new English voice, "Bubbie."
- Expanded the variety of short phrases the assistant uses to signal it is searching.
Why it matters: A more natural, multilingual voice assistant directly improves the experience for people relying on ChatGPT Voice in daily work and multilingual customer support.
2026-10-16 Morning edition
Trend Overview
- Efforts to demonstrate real-world AI safety continue, with independent verification frameworks taking shape.
- Competition to train foundation models remains fierce, with large in-house pretraining runs still underway.
- Market concerns about overheated AI investment and regulatory moves generated plenty of peripheral news, but this edition kept only the items meeting our source-independence standard.
2026-10-16 Morning edition
Summary
Top 3 Takeaways
- Anthropic gave outside evaluator METR staff-level access, strengthening its safety verification framework.
- OpenAI updated ChatGPT Voice to GPT-Live-1.1, improving multilingual speech naturalness.
- Only two confirmed new stories made today's cut, reflecting strict application of our source-independence standard.
What to Watch Next
- What findings METR publishes from its evaluations.
- How far GPT-Live-1.1's language coverage expands and how users respond.
- Ongoing developments in AI safety and regulation.