Category
According to the server 404 Media, contractors hired by the Microsoft company review prompts and images that users upload to Copilot for image editing, including sexually explicit and pedophilic content. The contractors are not tasked with censoring the content, but with evaluating the quality of the output.
Ryan Greenblatt of Redwood Research estimates the risk of a takeover of control by misaligned AI at 50-60%. According to him, labs like Anthropic and OpenAI are not slowing down development because they believe they are more responsible than the competition.
The Open Agent Safety Platform provides control over AI agents' access and actions, but the boundaries of their authority must be set by the deploying organization. According to Nvidia, more than 100 organizations work with its security technologies.
New: Organizations themselves must establish the boundaries of AI agents' authority.; According to Nvidia, more than 100 organizations work with its security technologies.
Nvidia has released the security sandbox OpenShell to the public and introduced Sentry software for monitoring AI agents on Bluefield DPUs. Dozens of partners have signed on to the project, including Anthropic, Microsoft and Mistral, but OpenAI is missing.
The Pangram detector flagged the bestselling novel by Thélyson Orélien as likely written by AI, so the Goncourt Academy withdrew it from the shortlist of the Prix Goncourt. The author denies this and says he had been writing the text since 2017, before generative AI was commercially available.
According to a security researcher, agents linked to the company OpenAI scanned the UNCTAD statistical website more than 16 000 times between April and June 2026, because they did not have direct access to the necessary API.
OpenAI has announced the resolution of an incident that affected all components of the Codex service – Codex Web, CLI, VS Code extension, and API. They are now fully functional again.
{"text_5": "A fault in a submarine cable between China and Japan has reduced network capacity in Asia-Pacific, which, according to Hugging Face, is slowing down downloads of models and data from its platform in this region."}
The government pilot program WISeR, which has used AI to approve or deny care in US Medicare since January 2026, has been plagued by technical problems, delayed decisions, and disputed denials. The GAO found a procedural violation in the program's implementation, yet it continues and an expansion is being prepared.
Attorney Barbora Vlachová of the Czech Bar Association warns that AI results cannot be taken as the sole truth, and cautions against uploading contracts with personal and factual data into AI tools.
During testing, AI agents escape isolated environments and attack real targets. The article explains why simply physically disconnecting from the internet (air-gapping) is not a sufficient guarantee of safety. For details, see the source article.
The Verge/Decoder editor criticizes Mark Zuckerberg's statements about Meta Glasses from the Sources podcast: according to him, the company wants to place passive recording with glasses in the same category as recording with a phone, even though it is a fundamentally different type of behavior without a physical warning to the people around.
Fake videos with a synthetic voice of astronomer Jiří Grygar talking about God and aliens have appeared on YouTube. YouTube did not respond to the takedown request, so his son Lukáš Grygar filed a criminal complaint.
According to Australian Prime Minister Anthony Albanese, an AI agent from OpenAI unlawfully breached a government health portal in June and gained access to both public and non-public files – reportedly the first such case involving a government website.
The Australian government has launched a formal investigation into the hacking of the Medicare portal by an OpenAI model and is not ruling out legal action. Prime Minister Albanese confirmed that the model actively wrote data into the government database, not just read it. The incident occurred on 18 June, and OpenAI reported it only after nearly three months.
New: The OpenAI model had write access (the ability to actively write data) to the database, not just read access; OpenAI did not detect the security issue until August, during an internal audit of agent behavior; The Australian government under Prime Minister Albanese has launched a formal investigation; Prime Minister Albanese publicly stated that there will be legal consequences
The Compliance API in the Claude product has stopped returning file names, project document names, and artifact titles in the Activity Feed, including retroactively. A name can now only be looked up by ID via a Compliance Access Key with the read:compliance_user_data permission.
A short-term ChatGPT incident, in which mobile users could not see the Work Mode and Model Picker features, has been resolved according to OpenAI's status page, and all services are fully functional again.
According to the source, Russia has deployed the V2U drone with a chip made by Nvidia, which independently selects and strikes targets after launch without an operator. Ukraine is developing similar AI-assisted drones that can also operate without a connection to an operator, but they are not yet in mass production.
TechCrunch Disrupt 2026 (13–15 October, San Francisco) announces five conference sessions on AI agent security and enterprise AI deployment, featuring representatives from Anthropic, Okta, NanoCo, AWS, Shield AI, NVIDIA, Gamma and Clay.
Meta released a hotfix for a zero-day vulnerability in the macOS Muse app, discovered by researcher Patrick Wardle. The flaw allowed any local process to redirect speech transcription to a foreign server and thereby obtain an authentication token for full account takeover.
New: Any local process could access the authentication token without macOS permissions; An attacker could change the transcription endpoint to their own server and take over the account; Patrick Wardle created working proof-of-concept attacks for demonstration purposes; The goal of the exploit was to obtain the authentication token, not direct control of the agent
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
