Google DeepMind and GovTech Singapore have fine-tuned a Gemma-based model to identify conversational scams while processing messages on the user’s device. Announced on October 14, the collaboration targets locally familiar deception patterns such as impersonation and investment or job scams. Deployment through applications will be phased; the announcement …
Liquid AI released LFM2.5-24B-A2B on October 13, pairing a mixture-of-experts architecture with reasoning and tool-use training for local agents. The model has 24 billion total parameters but activates two billion per token. Its context window grows to roughly 131,000 tokens, and both reasoning and base weights are available …
Google released Gemini 4 Argon on October 13 through developer APIs and several paid consumer services. The model is positioned for sustained coding, professional and scientific work, with introductory API prices and a combined input-output token budget. Google also reports internal engineering results from its use in Antigravity, …
Meta introduced Muse Wonder and Muse Wonder Full on October 12, extending its model lineup toward agents that retain context and work on longer tasks. Muse Wonder is available through the Meta Model API and Muse Code, while the Full variant is offered to Muse Code subscribers. Meta’s …
NVIDIA is contributing Vera Rubin rack and compute-tray reference designs through MGX to the Open Compute Project, alongside open-source tools for operating AI infrastructure. Announced on October 12, the work addresses the surrounding power, cooling, networking and assembly systems needed to build larger AI datacenters. The contribution concerns …
Microsoft says its Maia 200 accelerators are now serving production AI workloads, while the next-generation Maia 300 will begin deployment in 2027. The October 12 announcement sets out a broader plan to make Maia the primary inference system for Microsoft’s products and services, with multi-gigawatt expansion starting next …
Cloudflare expanded its open-weight decision-model family on 9 October with Clef-omni, adding audio and video to text and image inputs. The release also changes the cost and capacity of hosted Clef-flash and speeds up hosted Clef. It gives developers distinct options for multimodal decisions, cheaper short requests and …
The European Commission announced a special frontier-AI safety meeting on 9 October, saying its Scientific Panel had been investigating recent loss-of-control incidents. Its experts had prepared questions for the model developers with the AI Office, and were due to present safety and security recommendations. The release signals regulatory …
Axios reports OpenAI and Anthropic executives preparing for public and political fallout after a severe AI incident. Decrypt's follow-on coverage describes worst-case cyber disruption of banking, internet access, power or water, alongside preparations to explain events to Congress. The concern is how a technical failure or misuse could …
The Technology Innovation Institute (TII) introduced a 1.6-billion-parameter speech-recognition model focused on Arabic and the Emirati dialect. Falcon-ASR transcribes five languages using the same weights and supports word-level timestamps. A demo is available; API access and native applications are planned.
Cloudflare's Basin analytics platform is also data infrastructure for AI applications: the company links fast setup to AI coding-agent workflows, and customer Bobsled describes distributing AI-ready data through it. Basin combines event ingestion, Iceberg table management and SQL queries; its three components are generally available separately. R2 storage …
GitHub outlined a ModernBERT classifier developed with Microsoft's Applied Sciences group to detect unstructured credentials. Existing AI detection customers receive the model; its push-time use has a separate rollout and billing scope.
OpenAI’s 7 October announcement outlined College Planner within its ChatGPT AI assistant for teens, combining application planning with questions about requirements and next steps. Its initial design targets US students; the announcement describes it as forthcoming, alongside separate study-tool updates, not already available.
NVIDIA's October report summarizes competition research documented in September. Separate Nemotron specialists and inference workflows reached scores above IOI and IMO gold thresholds; the IOI run was unofficial.
Google introduced experimental Playground on 7 October. US adults can create browser games through text prompts and share them privately or publicly. Creation access is rolling out by Google AI subscription tier.
Google Research reported on an existing patent-lawyer trial that separates AI-assisted output from unassisted judgment. Drafting improved with tool access, but later error-correction gains were concentrated among experienced lawyers.
Liquid AI released d1-3B and the experimental d1-omni-600M on Hugging Face. The models answer structured questions in one forward pass: d1-3B accepts text and images, while d1-omni-600M accepts text with images or audio.
NVIDIA and Microsoft’s October 7 Windows AI announcement covered operating-system-managed background agents, RTX Spark PCs and a DGX Station for Windows preview. A follow-up NVIDIA announcement dated October 17 says RTX Spark laptops are now available, following earlier preorders and the previously announced October 16 availability target. Compact …
Google announced EmbeddingGemma 2 on October 6, a lightweight embedding model for retrieving information across several media types on consumer hardware.
OpenAI announced a staged ChatGPT rollout on October 7, combining GPT-6 with visual and interactive answers. Availability varies by tier and administrator settings.
Google Research published a workshop report on 5 October drawing on discussions among more than 50 academic and industry participants. It applies Contextual Integrity to agent information flows and actions, and proposes a supervisor layer with contextual policies. Research directions include dynamic sandboxing, model reasoning, user controls, multi-agent …
GitHub introduced ReviewBench on 5 October, with 219 public pull requests across 19 languages. Its reference findings combine human review, frontier models and static analysis, while evaluation distinguishes known findings from newly discovered valid issues. GitHub reports alignment with its production experiments. These results describe the benchmark and …
OpenAI announced that API customers worldwide can opt in to text watermarking for select models, with the API default remaining off. Eligible ChatGPT and Codex text output in the EU will be watermarked over the coming weeks. Detector access initially requires approval and is limited to researchers and …
OpenAI released GPT-Live-1 in its API for voice applications. The model can listen while speaking and keep a conversation moving while a separate backend handles reasoning and tools.
The European Commission opened a targeted copyright consultation on 29 September, with responses due by 3 November 2026. Its survey covers generative AI, licensing and performers alongside other copyright topics, in preparation for a possible initiative.
Google launched limited-access Fairwind on 2 September for government agencies, Google Cloud customers and cybersecurity partners. The offering combines Gemini 3.8 Flash Cyber with CodeMender to find, verify and fix vulnerabilities. Access is staged for trusted partners.
Google DeepMind released AlphaGenome Atlas on 8 September, making precomputed predictions for about nine billion possible single-letter DNA changes searchable. Its Variant Impact score helps rank variants for follow-up research. These are model predictions, not laboratory measurements or established clinical interpretations.
Google DeepMind published SynthID Bio, research methods for embedding detectable signatures into AI-generated protein sequences and predicted structures as a provenance signal. The publisher reports proof-of-concept wet-lab results and notes that tampering resistance and broader validation remain open.
The European Commission adopted the EU KIDS Act proposal on 17 September 2026; it is not enacted law. The Commission says it would prohibit social media platforms from granting access to children under 13, limit independent account creation below 15 on certain online social networking services and video-sharing …
OpenAI announced a public-beta Agents API that exposes its managed Codex agent harness to developers, covering session orchestration, context management, and tool use with a choice of execution environments.