A preprint is a signal, not a finished product or an independently confirmed result. We therefore track research papers separately and show the actual state of supporting evidence for each one.
1381
published research events
Latest work
A significant claim from a single source is published only after further confirmation.
The study introduces SRFT, a method that uses agent failures and analyses of those failures produced by an expert model. According to the authors, it reduces the success rate of prompt injection while maintaining task performance. The tests include both static and adaptive attacks; the code has been published.
The TrustMI study uses 2 000 paired conversations to steer assistants' trust without changing model parameters. Across six models from three families, the authors found that activation interventions change trust decisions in both directions; they also examined safety scenarios.
The study introduces the LMBuild benchmark and evaluates 30 systems on designing structures that can be assembled. According to the authors, functionality and physical usability remain challenging; specifying functional requirements improves part completeness, kinematics and physical usability.
The study introduces ADSD, a workflow that combines diagnosing numerical problems, finding suitable methods, and creating reusable skills. The authors report improvements in four areas and a nearly 71-fold reduction in average error when solving GOC-500.
The authors propose evaluating how AI agents work using five criteria: context sensitivity, temporal continuity, coordination across modalities, adaptive interaction, and monitoring their own execution. For each criterion, they describe measurable indicators and a testing procedure using controlled changes.
The study introduces SpaC-MEM, a memory system that links conversations to objects using 3D reconstruction. According to the authors, it achieves the highest accuracy among the evaluated methods while using fewer input tokens. The new Ego-SpaCR benchmark contains 620 video recordings, 95 conversations and 3 100 queries.
The study compared three language models. Two variants with fixed token codes eliminated 100.7 million trainable parameters and demonstrated an ability to solve tasks. However, the control model with learned input vectors performed better in several evaluations.
The study authors report that auxiliary tasks aligned with the input representation improved the prediction of optimal chess moves with the same amount of target data. Tasks capturing the dynamics of the environment provided the most consistent gains; the generation of factual commentary also improved.
The study introduces MS-Exam-Gen: 66 sources were used to create 3 058 test questions about MRI in multiple sclerosis. Across 12 LLM endpoints, accuracy ranged from 46.9–89.7 %. The test assesses textual knowledge, not image interpretation; results are affected by the order of answer options.
The study introduces the CLARA framework and a collection of 1 107 Chinese-English children's stories. According to the authors, structured annotations improved agreement with developmental reference standards and educators' assessments compared with readability-based methods and direct prompting.
The study compares directly optimizing predictions with learning the full latent representation for the next time step. In the four-dimensional example presented, the second approach has an error of 2.794 at rank 1, while direct optimization achieves 0.992. At rank 2, the results are identical.
The study introduces the RISED method: evaluations of agent behavior guide data selection, self-distillation and further generation. According to the authors, it achieves the highest average success rate across the tested environments; the available excerpt does not provide numerical results.
Google Research published a report from a workshop with more than 50 participants from academia and industry. It describes open questions about the privacy and security of AI agents and emphasizes that whether sharing data is appropriate depends on the context.
According to a survey by Quinnipiac University, 47 % of adults in the US want to slow AI development and 30 % want to halt it until its safety is verified. Independent safety standards are supported by 86 %. The survey included 1 202 respondents, with a margin of error of ±3.5 percentage points.
Three researchers argue in a paper published on 6 September on ArXiv that AI agents may overwhelm users with information and turn oversight into rote approval. The available excerpt does not provide experimental validation of this conclusion.
According to a survey of 300 executives, an average of 34 % of AI agent projects reach production. Data fragmentation was identified by 55 % of respondents as one of the main barriers to expanding agents’ access to knowledge.
The arXiv repository now allows a maximum of two papers per author per calendar month and three simultaneously active submissions. According to the operator, low-quality texts created using AI are overburdening volunteer moderators and delaying the review of high-quality papers.
Researchers at Wesleyan Media Project have recorded nearly 170 ads using AI in US campaigns in 2026, with spending of approximately 80 million USD. According to their analysis, Republican candidates and supporting groups account for 80 % of the ads and 83 % of the spending.
AI Radar monitors Czech and international sources every day, looking for changes that truly deserve attention.
MonitorsOfficial AI company blogs, specialist media, and research sources.
Selects and combinesFilters out information noise and combines articles about the same change into a single event.
Summarizes and explainsExplains significant events in English: what happened, why it matters and where the information comes from.
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.
Everything you need to navigate the AI world
Today’s briefingAn overview of news from the past 24 hours and a main story when available. Beside them is Live · AI Flash, followed by research and links to other Radar sections. On mobile, the blocks appear one below another.
AI FlashAn ongoing feed of brief updates with an evidence status. Links lead to a Radar detail page when one is ready, otherwise to the original source. You can also find reset and outage histories here.
Practical applicationsWhat new tools and features can do, what you can try and what their actual impact could be.
Model selectionModel comparison by type of work, capabilities, price and speed.
Research and archiveA separate research overview, topic search and older events by date.
One event, everything that matters
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.
Importance: ▮▮▮ majorOpenAIModels✓ 6
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.
▮▮▮ major · ▮▮ important · ▮ we're tracking = how significant the change is✓ 6 = six independent publishers, not the number of articles or feeds✓ official = a clear release, law or incident is substantiated by the relevant authority1 source = no independent confirmation yetbold = who is behind the changegray text = a brief summary of what happened
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.
The same news, two practical uses
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.
For individualsWhat you can use or try, how the change can help you at work and what to watch out for.
For businessesWhat impact the change could have on processes, costs, risks and other business decisions.
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.