MiMo

Xiaomi opens its desktop agent: MiMo Desktop enters invite-only beta with two new preview models

Xiaomi opened MiMo Desktop to invited testers. The app reads your own material, breaks the task down and returns editable files, alongside the new MiMo-X-Pro-Preview and MiMo-X-Flash-Preview models.

NeoMME

NeoMME: an open-source model that searches scanned documents without OCR

H Company released two encoders, Apache 2.0, that read a PDF page as an image and search inside it. The 260M version comes within a hair of a model 14 times its size.

Meta

Meta publishes alignment-free Text-Audiobox research

Meta AI has posted a paper on an alignment-free Text-Audiobox for voice dubbing and full-duplex dialogue synthesis; only the title is available so far.

OpenBMB

OpenBMB Ships MiniCPM5-2B, a 2.52B Dense Model Aimed at On-Device Agents

OpenBMB's Apache 2.0 licensed MiniCPM5-2B reports a 53.9 average across 34 benchmark rows, with its widest leads in tool calling and coding agents.

OpenAI

New Court Filings in the Times Copyright Case Against OpenAI and Microsoft

The New York Times argues in its lawsuit that OpenAI and Microsoft used publishers' copyrighted articles to train artificial intelligence systems.

Insilico Medicine

AI-designed lung drug shifted patients' biological age markers younger

Rentosertib, developed by Insilico Medicine with AI, moved biological age scores toward a younger state in trials, with one method showing a gap of about six years.

Amazon Bedrock

AWS shows how to tune Bedrock knowledge bases for complex documents with Textract

AWS published a reference solution pairing Amazon Textract with Bedrock knowledge bases to improve RAG accuracy on long, table-heavy documents such as utility bills.

Microsoft

Invisible Unicode tags: spammers adopt ASCII smuggling to slip past filters

Microsoft says ASCII smuggling, once used to hide prompt injections from AI agents, is now a spam tactic, with daily detections jumping from 21,000 to 2.5 million.

Z.ai

A refusal-stripped GLM-5.3 derivative is circulating, tagged for offensive security

The model on Hugging Face is built from the official GLM-5.3 weights with safety refusals removed. It carries pentest and offensive-security tags and has passed 15,000 downloads in a week.

CrowdStrike

A Berlin district blocks the US security tool brought in to clean up its hack

After the Rhysida attack, Berlin's Senate hired CrowdStrike for forensics. The Lichtenberg district refuses to run the software on its network, citing near unlimited access to staff machines.

White House

The US and China are preparing their first talks devoted only to AI

Reuters reports a dialogue planned for mid-September, led on the US side by Treasury Secretary Scott Bessent, with AI-directed cyberattacks on the agenda. A White House official said no such meeting is currently planned.

OpenAI

OpenAI chief scientist: no lab is in a position to keep scaling at full speed

In an essay on OpenAI's site, Jakub Pachocki calls for voluntary slowdowns and international coordination. He also says the company's main safety tool, chain-of-thought monitoring, is becoming less reliable.