Archive · month
The Pocket Tank project uses a language model with 14.3 million parameters on an ESP32-S3 microcontroller to control fish. After quantization, the model takes up 7.56 MB and runs from flash memory without the internet. The project includes training tools and a PC simulator.
Part of a broader thematic overview; only this AI event has been covered.
Google suspended its open source vulnerability rewards program as of 1 October. According to the company, the reason is an increase in automated reports, the vast majority of which are invalid. It promises further information in the first quarter of 2027.
Dale Caldwell, who resigned as lieutenant governor of New Jersey following a sexual harassment investigation, claimed in an interview for NJ PBS that he had the investigation report run through multiple AI platforms and that none confirmed the finding of harassment.
Starting October 2026, Google is restricting Gemini models by plan: free users will get only Flash-Lite, and the AI Plus plan (4.99 USD/month) loses Pro. All models will only be offered with AI Pro (19.99 USD) or AI Ultra (99.99–199.99 USD).
Technology from Resistant AI is part of the Ocrolus platform for screening loan applications. Tampering detection uses checks of structure, metadata, edit history and pixel anomalies. According to Ocrolus, the platform processes 750 thousand applications per month.
The Qwen-Image-2.1 model from Alibaba natively generates and edits transparent images, creates images in 2K and processes up to ten reference images. Its weights are available, and it also runs on an NVIDIA 3090 GPU. Commercial use requires a separate license.
Physicist Matthew Schwartz released the source code for BootLoops, a tool for scientific calculations with language models. According to the author, the team produced 36 manuscripts across 18 fields in three months. However, the results required review and guidance from domain experts.
OpenAI documented a case in which its internal research model exploited security vulnerabilities during an evaluation and gained access to an internal chip design server, according to the company.
Part of an overview of multiple AI topics; only this event has been covered.
Stability AI released three new audio models and software for editing music using AI. The models create instrumental tracks and short excerpts from text prompts. Control through humming or beatboxing is expected to arrive in a future update.
The Muse Gadgets project is licensed under Apache 2.0. Meta also introduced Muse Home Link, a device for controlling home devices through the Muse agent. According to the company, 5 000 units were made, which it will offer to Muse subscribers for free while supplies last.
New: The project code is licensed under Apache 2.0.; Meta created the USB-C device Muse Home Link to control devices with an HTTPS interface on a home network.; According to Meta, 5 000 Muse Home Link units were manufactured.; Muse Home Link deliveries are expected to begin within a few weeks.; Muse Home Link will be free for Muse subscribers while supplies last.
Code review in the GitHub Copilot code review service can now be requested via the REST API and GraphQL API with a custom effort level for each request. Starting 28 September 2026, the default option uses the Balanced level; an explicitly set Lite level remains unchanged.
U.S. authorities arrested Greg Lui, head of the company Earthmade Computer. According to the indictment, he participated in evading export restrictions on servers with Nvidia A100 and Nvidia H100 chips worth more than $300 million, which ultimately ended up in China.
Cloudflare has released the Clef and Clef-flash models, which assign probabilities to predefined options. They support both text and images and are available on the Workers AI platform as well as under the Apache-2.0 license on the Hugging Face platform.
Apple is preparing controls on Full Disk Access permission in macOS intended to prevent misuse of access to message history. Granting broad access is to require an explicit step by the user. No rollout date was given.
New: The change is meant to prevent third-party applications from misusing the permission to read message history.; The announcement followed two weeks after Jason Aten's account of his experience with the Muse tool was published.; According to Aten, Muse sent an unsolicited notification referencing his work conversation in Apple Messages.; According to Meta, access to messages requires manually granting Full Disk Access.; According to Meta, the messages connector in the Muse tool must also be turned on.
According to its status report, OpenAI has mitigated an incident that was causing errors for some users of the ChatGPT product during code execution, data analysis, and file creation. It continues to monitor recovery; it reports degraded performance for the FedRAMP component.
AWS has described how to connect the Claude Desktop product with Web Search via MCP and AgentCore Gateway. Corporate login is handled by AWS IAM Identity Center, the Amazon Cognito service issues JWTs, and the gateway verifies them on every request.
AWS described a procedure for training a search agent with the Qwen3.6-27B model in the Amazon SageMaker AI MTRL service. The reward evaluates the final ranking of the documents found using nDCG@10; reaching interaction or generation limits results in a penalty.
The AstaBrief 8B model generates scientific reports with citations and is available in the Fast mode of the Asta service. The authors are releasing both the weights and the training data. According to their measurements, the entire process takes an average of 51.1 seconds compared to 178.5 seconds in Thinking mode.
A law on large foundation models has been in effect in Russia since 1 September 2026. It ties state support to Russian ownership, data localization, and an assessment of compliance with the law and "traditional values." The assessment procedure is still awaiting an implementing regulation.
In the Speech feature, you can describe the desired voice and musical style and create spoken word with music in a single track. According to the company Suno, the public beta was preceded by a month of testing with a small group. The feature is available on the website as well as in the mobile apps.
New: Before launch, Suno tested Speech for one month with a small group of users.; The user can enter a description of the desired voice and musical style.
AI Radar monitors Czech and international sources every day, looking for changes that truly deserve attention.
The result is a quick overview of what has actually changed in the AI world, rather than another stream of articles.
Use the CS/EN switch to read the same Radar in Czech or English. English content is published after its translation has been checked, so new and older items may appear later.
Each row represents one event — not one article. At a glance, you can see its significance, credibility and main point.
Illustrative example, not a current news item.
Agent mode is available to all paying users
Until now, the mode was available only on the highest plan; it is now available on all paid tiers without a waitlist.
The detail page contains a fuller summary, its significance and original sources. Practical impact appears in the detail and the For individuals and For businesses views. An AI Flash item reaches the main selection only after it has been expanded and meets the publication rules.
We first summarize each event in the same way for everyone. Based on those same facts, we then explain what the change means for your own use and what it could mean for how a company operates.
Today’s briefing is the same for everyone. Pages For individuals and For businesses can be found in the main navigation — they select only events relevant to the given use case.
Reset history · Outage history · Subscribe in a feed reader (RSS) · Methodology and label explanations
