Skip to content

AI Flash · live feed

What is happening in AI right now

A timeline of everything that has passed through the radar — newest first, with an evidence status for each item.

2524 across the entire feed ↗ details, original sources and our own analyses

Continuously updated

Latest items

Filter contains 291 items

How to read evidence statuses ↓

Yesterday

1 item

Friday, November 20

7 items
Security clearly official source

GitHub Enterprise Server 3.23 (release candidate) adds local AI secret detection

GitHub has released the GHES 3.23 release candidate with a new AI secret detection feature. According to the company, it recognizes unstructured secret data directly on the appliance without a GPU, a hosted LLM, or an internet connection, including in air-gapped environments. The feature is in public preview.

Part of a broader thematic overview; only this AI event has been covered.

GitHub Changelog (Copilot and AI features) Details and sources →

Thursday, November 19

7 items
Tools and apps clearly official source

The company Amazon has published an independent validation of AI-conducted interviews in the Amazon Connect Talent tool

The company Amazon has published an independent assessment of the accuracy of AI-conducted interviews in the Amazon Connect Talent tool (launched in September 2026). According to Amazon, the AI's assessment agreement with experts reached 89.3%, 97.6% for clear evidence, and Cohen's kappa was 0.79.

AWS Machine Learning Blog Details and sources →

Wednesday, November 18

6 items
New models clearly official source

GitHub Copilot has made the Gemini 4 Neon model from Google available

GitHub Copilot has newly made available the Gemini 4 Neon model from Google for agentic programming. According to the manufacturer, it is cheaper per token, faster, and makes fewer unsolicited changes than the Gemini 4 Carbon model. Available for Pro+, Max, Business, and Enterprise, with a gradual rollout.

GitHub Copilot Changelog Details and sources →
New models clearly official source

RadixArk releases open-weight model DaisyRose-27B for agentic tasks

RadixArk has released the DaisyRose-27B model under the Apache 2.0 license, a fine-tuned version of the Qwen3.8-27B model focused on long agentic tasks. On the Terminal Bench 4.1 benchmark it improved pass@1 from 5.6% to 12.0%, and other coding and instruction-following tests also showed improvement.

Hugging Face Blog Details and sources →

Tuesday, November 17

3 items