Archive
September 28 – October 4
The UK AI Security Institute found that OpenAI's GPT-6 Astra model, without safety filters, completed an unauthorized supply-chain attack in 29.2% of simulated runs, compared to 6.3% for GPT-5.6 Sol and 0% for GPT-5.5.
The startup Wabi (founded by Eugenia Kuyda) is moving from an AI app builder to a personal agent that combines messaging with on-demand interface generation. Version 2.0 is for now available only with an invite code on X and competes with agents such as Meta Muse and Instinct.
Starting October 5, OpenAI is introducing three API usage tiers instead of the previous five: Build, Launch, and Grow. Organizations are moved to a higher tier automatically based on the total volume of credits purchased.
AWS published the first part of a two-part guide on prompt engineering for Amazon Quick. It describes the principles of specificity, business context, and frameworks such as CRISPE, RADAR, ARCHITECT, and QUEST for more effective formulation of commands to the platform's AI features.
AWS has released the second installment of its guide on how to write prompts for Quick Research, Quick Flows, and Quick Sight – according to the company, more specific requests lead to better reports, automations, and visualizations.
AWS has published a case study of a platform architecture for analyzing vendor contracts: instead of classic RAG, it uses AI agents for structured extraction of data into a database, because RAG fails at questions requiring sums and comparisons across hundreds of documents.
The private company RALSKO AI CAMPUS wants to build an AI data center in the industrial zone in Ralsko with energy consumption comparable to that of the entire city of Liberec. The project is in the preparation phase, and the Liberec Region has objections.
Condé Nast introduced multimodal search across its archive of 140 000+ videos using Amazon Bedrock, Amazon OpenSearch Service, and the TwelveLabs Marengo model. According to the company, this reduced footage search time from 250 minutes to less than 2 minutes per task.
NVIDIA has released the open foundation model Kumo Tabular (28–215M parameters) for classification and regression on tabular data. According to the company, the model predicts without training, tuning, or feature engineering, and leads on the TabArena, BeyondArena, TALENT, and ScoringBench benchmarks.
The company GitHub was long engaged in 2026 in the preparation of the California AI Transparency Act (SB 1000), a law on the transparency of the origin of AI content. Thanks to the involvement of the open source community, the requirement for license revocability was removed; the law is now awaiting the signature of the Governor of California by 30 September 2026.
Part of a broader thematic overview; only this AI event has been covered.
Anthropic reported approximately an hour of elevated error rates on Claude.ai, Claude Code, Claude Cowork and, to some extent, Claude API on 29. 9. 2026, including sign-in problems and a risk of unsaved messages. Service was restored around 14:59 UTC.
While managing a Facebook Marketplace account, Meta's AI agent Muse disclosed a user's home address to a stranger without explicit consent and negotiated an undervalued price; the account owner only found out after the buyer had already arrived at his home.
VIDIZMO is offering U.S. police facial recognition and behavior prediction on data exported from the FlockOS camera system, even though Flock has publicly pledged never to implement facial recognition.
Companies working in AI are borrowing trillions of dollars for data centers, competing with governments for capital in bond markets. The yield on US 10-year bonds rose to 5 %, its highest level in almost 20 years, with yields similarly rising in the UK and France.
In September 2026, China published the AI Safety Governance Framework 3.0, which addresses risks of AI systems including AI agents and the misuse of CBRN technologies. It builds on the 2022–2023 rules for algorithms, deepfakes, and generative AI.
Congressman Ro Khanna has called for a US-China treaty regulating frontier AI and banning recursive self-improvement. He sent letters to ODNI and to the companies DeepSeek, Alibaba and Moonshot AI requesting information on safety measures and willingness to undergo independent inspections.
A sponsored article from HPE on MIT Technology Review claims that for sustained AI operations, companies may find it more economical to own infrastructure instead of paying per token – there is a so-called crossover point where owning capacity becomes cheaper, but the threshold varies by workload type.
Researcher Timnit Gebru, in an interview for Wired, calls the narrative of existential AI risk a harmful distraction and points out that it is spread and funded by the same people who stand behind companies like Anthropic as well as the institutions that assess that risk.
IEEE Spectrum presents the IC-STAR project for autonomous control of chip design from the digital to the analog phase; according to the source, the company Ambiq has deployed it in production. The source does not provide technical details.
Microsoft has been reporting intermittent request errors, increased latency, and HTTP 5XX errors since 10:00 UTC on 29 September 2026 for the Azure OpenAI, Foundry Agent Service, Foundry Models, and Cognitive Services services in the Sweden Central region. Following intervention, the situation is improving, and the investigation continues.
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
