Track model releases, AI agents, open-source projects, infrastructure, policy, and product updates as they publish.
AI lab Hark has released a personal assistant that puts privacy front and center, with TechCrunch describing it as an 'operating system from the future.' The product is positioned to compete with rivals including Muse, Dots and Instinct.
AlphaSense has launched SuperAnalyst, an AI agent positioned as an always-on assistant for financial and strategic workflows. According to Dealroom, the emphasis is on continuous, proactive operation rather than a passive tool that waits for each query.
Mirror Particle is building a ‘world model’ of human behavior. Mirror Particle will launch at TechCrunch Disrupt's Startup Battlefield 200 with a world model built from scratch to predict human behavior, arguing that LLM role play falls short for market research and brand strategy.
OpenAI and Atlassian are expanding their partnership to connect frontier models with enterprise knowledge so teams can plan, build, and deliver work. The official announcement points to models reasoning with an organization's own context rather than serving as a generic assistant.
Nvidia has outlined a security framework for AI agents and highlighted OpenShell, its open-source runtime for enforcing controls outside the agent's reasoning. The framework argues security depends on the full agent stack rather than the model alone, applying identity, access control and verifiable safeguards to systems that reason, use tools and adapt their actions.
Anthropic announced an expansion of its Claude for Startups program, offering qualifying companies a free year of Claude Team with up to five premium seats plus $1,000 in API credits. Selected firms also get access to Claude Marketplace for building plugins and can book virtual office hours with Anthropic's Applied AI team.
China's DeepSeek and its peers launched 16 AI models within a single month, according to Nikkei Asia, a pace the report frames as continuing despite a warning from Anthropic. The density of releases underscores a high-frequency contest among Chinese developers and keeps friction over model safety and usage boundaries in focus.
The maker of the open-source LibreOffice document editor says it has no plans to add AI to the software's default configuration, citing user privacy. The stance turns the absence of AI into a deliberate selling point at a time when rival productivity suites are racing to embed assistants.
Google Research has published a blog post describing how it is unlocking Earth AI's planetary-scale geospatial foundation models for use in global public health. The work aims to turn Earth-observation data and foundation-model capabilities into a base layer for health applications such as disease surveillance and environmental exposure assessment.
French AI lab Mistral AI has released Mistral Large 4, a new large multimodal model with around 1 trillion parameters. The company frames it as a bid to leapfrog rivals in both the United States and China, across both closed and open model camps.
Pinterest has introduced AI-powered Beauty Guides that turn saved hair and nail Pins into actionable beauty plans. The feature translates the styles into salon terminology and adds estimated costs, appointment times and maintenance needs, moving users from inspiration to a concrete booking decision.
Time reports that Meta's Muse AI agent is building a dossier on its users. The coverage puts the intersection of AI agents and user profiling under fresh scrutiny.
Netskope has launched an AI agent marketplace aimed at security teams, giving them a single place to deploy its AI tools for security and network work instead of buying each one separately. The move is part of a broader push by security vendors to turn agentic AI capabilities into packaged, purchasable products.
NVIDIA announced a 30-billion-parameter Nemotron 3 Large Telco Model, fine-tuned by AdaptKey on open source telecom datasets to handle tasks such as network configuration and customer incident triage, and released the full end-to-end fine-tuning recipe. Its State of AI in Telecommunications report found that 89% of respondents consider open source models and software important to their company's AI strategy.
CXApp has launched an AI agent aimed at automating enterprise workflows, moving agentic AI from chat interfaces into day-to-day business processes. Details are still limited, but the release fits a broader push by enterprise software vendors to embed agents in routine tasks.
OpenAI published new results on open problems in mathematics produced by an internal frontier model. Alongside the findings, the company released 722 math manuscripts and shared Lean proof formalizations and research details on GitHub.
OpenAI published a customer story on how quantitative research firm Jump Trading is scaling its quant research with ChatGPT. According to OpenAI, the firm uses longer-running AI workflows that combine multiple data sources with human review.
BioRender has launched Leo, described as the first AI agent built specifically for creating scientific figures, letting researchers sketch, build and edit diagrams through conversation. Rather than returning a single flat image, Leo asks focused questions, produces a sketch to mark up, and turns that into an editable figure on BioRender's canvas.
A report from nokiamob.net says free users of Google's Gemini assistant will lose access to the Flash and Pro models. The change narrows the model lineup available at no cost, lowering the capability ceiling for free accounts.
A study by FTA Global has scored Indian e-commerce websites at an average of 43 out of 100 on AI Agent Readiness. The result points to clear gaps in how well local shopping platforms are prepared for AI agents.
Descartes has launched a product called the AI Agent Control Plane at its Innovation Forum, according to a StreetInsider report. The launch targets a growing enterprise need: governing and coordinating fleets of AI agents under one managed layer.
OpenAI and the contracting software company Ironclad are partnering to train and evaluate AI agents on complex contracting workflows, aiming to push computer use beyond demos into professional work. The collaboration focuses on whether agents can operate real software reliably enough for enterprise contracting tasks.
Egypt Oil & Gas reports that ERC is trialing an AI agent to predict equipment failures, shifting maintenance work from reactive repair toward early warning. The deployment is a representative example of industrial agents and shows how energy firms are treating predictive maintenance as an early AI priority.
Media reports say a security vulnerability was discovered in Meta's Muse AI agent before it launched and was quickly patched. The reports do not disclose the technical details, but the episode puts pre-launch security review for agentic products back in the spotlight.
An official plugin called MCP Adapter has arrived in the WordPress directory, letting AI agents connect to WordPress-powered websites. The release brings the MCP standard for linking agents to tools and data into one of the web's most widely used publishing platforms.
Codex lead Tibo's 28-day plan opened with a speed upgrade, lifting default inference speed for GPT-6 Astra and GPT-6.1 Sol by roughly 50% to 50 TPS. The daily delivery-or-quota-reset pledge is meant to win back users frustrated after GPT-6.1 Sol ran slowly under heavy load.
UAE-based TII has published Falcon-Emirati, a large language model built to capture the Emirati dialect, local culture, and conversational nuance. The release points to a growing push to make multilingual models fluent in regional variants rather than only standard forms of a language.
According to a report from Sina, Chinese AI model company Moonshot AI has completed its pre-IPO funding and plans to list in Hong Kong in the first quarter of next year. If the listing proceeds as planned, it would mark a significant step for a leading Chinese AI startup in tapping public capital markets.
A report surfaced through Google News CN highlights Jev, said to process trillions of tokens a day, as the spark for a new AI track around "decision models." Silicon Valley firms are described as collectively positioning themselves in this direction.
The U.S. Federal Trade Commission is examining OpenAI and Anthropic over safety risks tied to their AI agents, according to a TechRepublic report. The scrutiny signals that regulators are turning their attention to autonomous agent systems as they move from chatbots into real business workflows.
Meta and Microsoft are reportedly reducing their reliance on Anthropic, shifting more weight toward their own internal AI tools. The report frames the change as a directional adjustment in how the two companies source and use AI rather than a product launch.
China's Roewe Jia Yue 07 has become the first production model equipped with the Doubao cockpit assistant, and its pre-orders have passed 30,000, according to Pacific Auto. The milestone puts a large-model in-car assistant into the sales process of a new vehicle rather than a concept or demo.
At a New York City Council hearing, recently departed staff from Anthropic, OpenAI and Google DeepMind described a reckless culture at the largest AI firms and warned that humanity is more likely than not to lose control of the technology. A separate panel of company staffers stressed a focus on safety as the Council weighs several AI regulations.
OpenAI will add an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act's transparency rules. The marks travel with copied text but can be weakened by editing, so OpenAI is limiting detector access to vetted researchers for now.
Meta has released the Muse Gadgets SDK, giving developers a way to build custom hardware around its Muse AI with screens, buttons and connections to other devices in the home. The move extends Muse from a software assistant toward a third-party hardware ecosystem.
According to The Next Web, OpenAI's own models were used to reshape its Jalapeño chip, with some of the resulting changes the company could not fully explain. The detail points to AI moving into chip design while exposing a lingering interpretability gap.
A report surfaced through Google News says NVIDIA is betting on what is described as a 'U.S. version of DeepSeek' and plans to release open-weight models. The move signals that the chip giant is pushing further into the model ecosystem and into competition with Chinese open-weight efforts.
California has issued a subpoena to OpenAI over cybersecurity breaches involving AI agents, according to a Security Boulevard report. The action signals growing regulatory scrutiny of how autonomous agents behave and the security risks they can introduce.
Sam Altman said the world must accept AI's 'bad things' as industry leaders signed a safety pact associated with President Trump, according to Fortune. The remarks pair an acknowledgment of AI risk with a coordinated industry commitment on safety.
Nvidia-backed Reflection AI has unveiled Beam, its first open-weight model, which the company says rivals GLM-5.2 on reasoning while requiring far less inference compute. The company says the weights will be released this month.
Daily Archive