Most safety checks on frontier AI models have so far been brief. An outside group gets access shortly before launch, runs its tests, and the model ships. That approach can catch obvious problems. It is less useful for judging whether a company's safety claims hold up across training, testing and real-world deployment.
OpenAI now wants to change that picture. In a framework titled "Priorities and principles for effective third party assessments," the company sets out how independent reviewers could get far deeper access to the way its models are built, evaluated and used. The idea is that assessors verify safety claims themselves rather than taking the company's word for it, and form their own view on whether safeguards actually work.
Voice assistants have mostly been good at conversation: ask a question, get an answer, move on. That works for looking something up. It is less useful for someone who wants to check their calendar and then act on what they find.
OpenAI is now trying to close that gap. In an announcement on X, the company said ChatGPT Voice can do more than respond. Spoken requests can now reach connected services and start work inside ChatGPT Work, the company's environment for producing documents and handling longer tasks.
Plugins bring connected services into the conversation
The main change is plugin support. ChatGPT Voice now works with plugins for email, calendar and Slack, so users can bring those connected services into a spoken conversation rather than switching to a keyboard.
Most of the debate about AI-generated text has focused on what these systems can write. Less attention goes to a quieter question: whether anyone can later tell that a machine wrote it. Anthropic has now given its answer for a larger part of its Claude lineup.
From 30 September 2026, the company will apply its previously announced text watermark to three more models: Claude Fable 5, Claude Sonnet 5 and Claude Opus 4.8. Anthropic shared the news with customers by email. Its stated reason is compliance with the transparency requirements of the EU AI Act.
Google DeepMind's new leader has a clear priority, and it is not artificial general intelligence. Koray Kavukcuoglu wants Gemini 4 out well before the end of the year. He has also made it plain that he sees little value in arguing over whether AGI has already arrived.
Kavukcuoglu made his first media appearance as head of Google DeepMind at The Information's AI Agenda Live Summit. There, he described Gemini 4 as being in an early phase of post-training, the stage where a base model is refined so that it behaves reliably. He said he hopes to release it considerably earlier than year-end.
Most of the talk about shrinking language models centers on 4-bit and 8-bit quantization. Ternarisation goes much further. It cuts every weight in a model down to one of three values: -1, 0, or +1.
That idea has been circulating in AI research for a while, most visibly through Microsoft Research's BitNet b1.58 work. It gets especially interesting at the scale of a model with around 1.85 billion parameters. A model that size is small enough to experiment with on modest hardware, but large enough to be useful for real tasks.
What ternarisation means
In a standard model, each weight is stored as a 16-bit or 32-bit floating-point number. Ternarisation replaces each weight with -1, 0, or +1, usually paired with a scaling factor that is shared across a group of weights or a whole layer.
CyrioX is financed by advertising. You can choose how you want to use this website:
With advertising: we load an advertising script from a third-party ad network. The ad network may set cookies, use your IP address and device information, and may process data outside the EU.
Ad-free for €0.99 per month: no advertising and no advertising tracking. Cancel at any time.
You can change your decision at any time via "Cookie Settings" at the bottom of every page.
