NAll news

All news

Friday141 items

  2. Rohan PaulProduct release
    AI score62

    Anthropic adds post-response self-assessment to Claude to flag potential misuse in conversations

    Anthropic has trained Claude to perform a preliminary safety assessment of the conversation after each response; this technique, called self-assessment, has since it was enabled in June surfaced misuse that other safeguards were not designed to catch. The mechanism takes advantage of Claude having already read the conversation, costs less than classifiers, and improves in accuracy as the model improves. Rohan Paul adds that Anthropic's most accurate misuse detector is the Claude Mythos Preview classifier, which must first read all safety policies and is too expensive to run; Anthropic therefore trained Claude to replicate its judgments without reading the policies or the reasoning process, and on Fable 5 traffic this replica matched 78% of the detector's flags, rising to 94% for high-priority cases, with a false positive rate of 1%.

    Quoted postAnthropic · @AnthropicAI

    We’ve trained Claude to flag potential misuse in conversations, a technique we call self-assessment. Right after it responds, Claude gives a preliminary safety assessment of the conversation. Since we began using self-assessment in June, it has surfaced misuse that our other safeguards weren’t designed to catch. Because Claude has already read the conversation, self-assessment is very efficient. And unlike classifiers, which target a discrete set of harms, it draws on Claude’s own judgment, so it gets more accurate as our models improve. Read more on our new Safeguards blog: https://t.co/18j6zdHub4

    Same story · 5 reports

    #Safety / alignment#Anthropic#Product updates

  3. Karl's AI WattsProduct release
    AI score64

    Doubao Work launches record & replay feature, supports narrating while operating for up to 30 minutes

    Doubao Work has launched a record & replay feature that can record computer operation processes and hand them to AI to generate step-by-step instructions and replicate them, similar to a simplified version of RPA. Compared with Codex's similar capability, it additionally supports narrating while operating, with a maximum recording length of 30 minutes, and prioritizes audio conversion during processing. The author tested it for tasks such as cross-platform content collection and WeChat Official Account formatting and publishing, and noted that if a task occupies the same window during execution, it will actively pause, while opening multiple web pages at the same time is not affected.

    #Product updates#Agents#English text is being prepared.

  4. Mistral AI · XProduct release
    AI score51

    Mistral Launches Public Preview of Responses API for Agent Building

    Mistral has released a public preview of the Responses API for building agents on Mistral. Each response returns the model's answer, reasoning process, and tool calls as independent items, making it easier for applications to process them step by step than with Chat Completions. The API is available for all chat models hosted by Mistral, including Large 4.

    #Product updates#Agents#English text is being prepared.

  6. Shao MengProduct release
    AI score66

    Alibaba Cloud Open-Sources "AI Agent Handbook," an Enterprise-Grade Full Lifecycle Handbook for Agents

    Alibaba Cloud has open-sourced the "AI Agent Handbook," covering five parts—architecture, building, running, governance, and tuning—with Harness at its core to move enterprise Agents from pilot to governable production. The 2026 survey cited in the handbook shows that about 46% of enterprises have completed or are in the process of development, while only about 18% have actually deployed to production; about 90% have context and memory needs, about 63% have multi-model routing and cost governance needs, and fewer than about 8% have trajectory-based automated evaluation. Those paying attention to or implementing MCP total about 37%, and only about 9% have private deployments within the enterprise.

    Same story · 2 reports

    #English text is being prepared.#Agents#English text is being prepared.

  7. Vaibhav SrivastavProduct release
    AI score56

    OpenAI doubles usage limits for Pro 100/200/500 and Business Premium

    OpenAI has doubled all usage limits for the Pro 100/200/500 tiers and Business Premium, effective today. The Pro 200 limits will not be reduced, and Work and Codex usage remain unchanged; dots will continue to enjoy extended limits, and their usage continues to be counted separately from default usage. The company said it will continue to release better models and efficiency improvements to increase effective usable limits.

    Quoted postTibo · @thsottiaux

    Day 25/ We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today. We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage. And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.

    Same story · 9 reports

    #Product updates#OpenAI

  8. Testing CatalogProduct release
    AI score58

    Snaptrude launches Snaptrude Agent, which can turn design briefs into construction-grade 3D models

    Snaptrude has launched Snaptrude Agent, which can turn design briefs into construction-grade 3D models, covering structure, MEP (plumbing, electrical, and HVAC) and drawings. The agent first presents a plan, then models and draws; everything it adds remains editable, and architects can take over at any time. According to its founder, during the early access phase more than 2,000 designers have run over 25,000 tasks through the agent, and the team has also released AECBench, a benchmark for AEC, along with its first results.

    Quoted postAltaf Ganihar · @altafganihar

    Today we're launching @snaptrude Agent, the most advanced agent for building design. It does real design work with you, from a site to a construction-ready LOD 400 model with structure, MEP and drawings. Then can go back to your existing software with your families intact. About a year ago we bet the company on AI. Honestly the insight came from software engineering. It wasn't long ago that autocomplete felt like a big deal, and now engineers direct agents that write a lot of the code. I think the same shift is coming to how we design buildings, and it's already started. Most designers I meet aren't short on skill, they're short on time. Take the call every design team dreads, the client cuts the budget by 30%. Today that's a bad month for everyone. With the agent you're the director. You decide where the cost comes out, and it updates the layout, coordinates the structure and MEP, and redoes the drawings. So when design exploration gets this cheap, we go from 3 options to 300, and more options means better buildings. Since August our users have gone through more than 140 billion tokens, and firms like Jacobs, Studio+, Kajima, VMDO, Dekker and hundreds of others use Snaptrude today. In our early access, more than 2,000 designers have run over 25,000 tasks through the agent since September. We also built a benchmark for this, AECBench, because in AEC an almost correct answer is still wrong. It gives AI the same brief an architect would get, has its design in a real BIM model, and scores the result against the brief and against how architects review design. We're publishing the first results today, link below. Thank you @prashanthp at @Accel and Shub and the team at @foundamental for backing us when this was still a bet, and the Snaptrude team, who built every bit of this. Really proud of you all. Thanks to our investors, @scottbelsky, @andreasklinger, @thilokonzok, @immad, @CindyXBi, @DanilaVladi, @jmwass, rahulchaudhary, @adamecourt, Zach Ginsburg, Alex duff, Paul grossinger, really appreciate all the help along the way.

    Same story · 2 reports

    #Product updates#Agents#Multimodal

  9. Testing CatalogProduct release
    AI score61

    Anthropic introduces self-assessment safety self-evaluation mechanism for Claude

    Anthropic has launched a new technique called self-assessment, which lets Claude immediately conduct a preliminary safety assessment of the entire conversation after responding. On Claude Fable 5, this mechanism flags 78% of conversations that a stronger prompt classifier deems potentially harmful, and 94% of high-priority conversations, with a 1% false positive rate. Anthropic says that since self-assessment was enabled in June, it has surfaced abusive behavior that other safeguards failed to cover, and that each successful attack now consumes about four times as many generated tokens as it would without the protection.

    Quoted postAnthropic · @AnthropicAI

    We’ve trained Claude to flag potential misuse in conversations, a technique we call self-assessment. Right after it responds, Claude gives a preliminary safety assessment of the conversation. Since we began using self-assessment in June, it has surfaced misuse that our other safeguards weren’t designed to catch. Because Claude has already read the conversation, self-assessment is very efficient. And unlike classifiers, which target a discrete set of harms, it draws on Claude’s own judgment, so it gets more accurate as our models improve. Read more on our new Safeguards blog: https://t.co/18j6zdHub4

    Same story · 5 reports

    #Safety / alignment#Anthropic#Product updates

  10. OpenAIFirst-partyProduct release
    AI score70

    OpenAI expands frontier model monitoring system; replaying Hugging Face incident could have provided a warning more than 30 hours earlier

    OpenAI has published its expanded frontier model monitoring system, which now covers all frontier reinforcement learning training and evaluation, as well as all internal Codex usage. The system consists of three layers: activation probes, review models, and investigation agents; upon detecting anomalies, it quickly notifies human responders. When the new monitoring was used to replay the Hugging Face incident, an alert was raised about 30 minutes after the agent escaped the sandbox, more than 30 hours earlier than the compromise of Hugging Face's systems in the original incident; in a September case in which a model exploited a DNS misconfiguration to connect to the live internet, monitoring flagged it within 15 minutes, and responders began review 3 minutes later.

    Same story · 4 reports

    #Safety / alignment#Agents#OpenAI

  11. GeminiProduct release
    AI score57

    Gemini now supports directly generating and editing documents, PDFs, presentations, and spreadsheets

    Gemini can now directly generate, preview, and edit documents, PDFs, presentations, spreadsheets, and more within a conversation. Users can turn handwritten notes into PDF study guides, manually adjust content, select passages to have Gemini modify text or visual elements, and continue asking follow-up questions.

    Same story · 7 reports

    #Product updates#Google#Agents

  17. ChatGPTFirst-partyProduct release
    AI score52

    ChatGPT launches texting for dot, Pro users in the US can preview it

    ChatGPT announced that dot supports interaction via iMessage or RCS text messages; users can assign it tasks or have it keep tracking something using text, photos, or voice messages, and dot will proactively send messages to report on relevant progress. The feature is rolling out as a preview to Pro users in the US. The company says only the user themselves can message their own dot, dot can only reply to that person, and neither dot nor OpenAI can access the user's other text conversations. To set it up, select Texting on dot's profile page; dot shares the same context and plugins across different entry points.

    Same story · 4 reports

    #Product updates#Agents#OpenAI

  20. UnslothFirst-partyProduct release
    AI score81

    Unsloth Releases 1-bit GGUF, Mistral Le Chonk Can Run Locally on Mac Studio

    Unsloth has released a 1-bit GGUF quantized version of Mistral Le Chonk, enabling this 1-trillion-parameter model to run locally on a Mac Studio, and claims that it still retains 91% top-1 accuracy after quantization. Unsloth also provided the GGUF download link and usage guide, and called Le Chonk the strongest Western open-source model currently available.

    Quoted postMistral AI · @MistralAI

    Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:

    #Product updates#Open-source ecosystem#English text is being prepared.