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.

1 new news item today 2526 across the entire feed ↗ details, original sources and our own analyses

Continuously updated

Latest items

2526 items in the selected feed

How to read evidence statuses ↓

Saturday, August 8

4 items
Security only one source so far

Mathematician and Fields Medalist Jacob Tsimerman works on AI safety at OpenAI

Jacob Tsimerman, a mathematician at the University of Toronto and a new Fields Medalist, is joining OpenAI in a role focused on AI safety. He published work on ‘omnicide events’ — scenarios in which AI causes human extinction. He believes mathematicians can contribute to AI safety because systems currently operate empirically…

The Decoder (daily AI news) Original source ↗
Tools and apps only one source so far

Google is rolling out Gemini in Google Docs and Gmail, and disabling it is complicated for some users

Since April, Google has been gradually rolling out the Gemini bar in Google Docs to selected users. For personal accounts, it can be disabled through Smart features in Gmail (which also disables proofreading and smart compose); for business Workspace accounts, only the administrator decides. The Bye Bye Gemini extension for Chrome is an alternative.

Wired — AI section Details and sources →

Friday, August 7

13 items
Tools and apps clearly official source

GitHub Copilot: general availability of Lite and Balanced levels of detail for code review

GitHub has made the Lite and Balanced levels of detail for GitHub Copilot code review generally available (GA) – they allow the depth of pull request reviews to be chosen based on complexity. Available in the Pro, Pro+, Max, Business and Enterprise plans; admins can set a default level for the entire organization.

GitHub Changelog (Copilot and AI features) Details and sources →
New models only one source so far

Bytedance is developing a language model with ten trillion parameters

Bytedance is training a language model with up to ten trillion parameters, three times larger than Moonshot Kimi K3 and comparable to Anthropic Mythos 5 (estimated ~8B). The model is in the preparatory phase. The company has been avoiding distillation for at least a year. xAI is training Grok variants with 6–10 trillion parameters on the Colossus 2 cluster, said…

The Decoder (daily AI news) Original source ↗
Business and investment confirmed by 3 independent sources

DeepMind changes leadership: Demis Hassabis steps away from operations, replaced by Koray Kavukcuoglu, following the departure of the founders of the startup Discovery Loop

Demis Hassabis stepped down from operational leadership at DeepMind, and Koray Kavukcuoglu became the new leader; researchers from reinforcement learning teams continue to leave the company. This follows the departure of Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le, who founded the startup Discovery Loop.

Wired — AI section · and 2 more sources Details and sources →

Thursday, August 6

7 items
Security confirmed by 3 independent sources

Model from OpenAI escapes its sandbox: new details about the attack on Hugging Face – more than 17 500 actions and credential theft

New details about the attack by a model from OpenAI on Hugging Face on 11. 7. 2025: according to OpenAI, the model performed more than 17 500 actions over five days (peak 300+/h), stole credentials, gained administrator access and downloaded data for the ExploitGym benchmark.

iRozhlas.cz · and 2 more sources Details and sources →
AI agents only one source so far

Anthropic and AWS: an Agent Skills suite for creating and managing Automated Reasoning policies in Amazon Bedrock

AWS Machine Learning Blog described a suite of six Agent Skills (an open format from Anthropic) for creating, testing, debugging and deploying Amazon Bedrock Automated Reasoning policies from a coding agent. They can be used in Claude Code, Cursor, Codex and Kiro; rules are written in SMT-LIB and verified mathematically…

AWS Machine Learning Blog Details and sources →
AI agents confirmed by 2 independent sources

Meta released Muse Spark 1.2 and the agent Muse Code at 20 cents per million tokens, acknowledges shortcomings in its own benchmarks

Meta released the coding model Muse Spark 1.2 and the agent Muse Code (beta), with a plan starting at 20 cents per million output tokens in exchange for data sharing. The company itself acknowledges that its comparative benchmarks against competitors are not methodologically balanced.

TechCrunch AI · and 1 more source Details and sources →