Catch today's product, model, and platform updates in a few minutes.
OpenAI has launched a new blog column, Intelligence Age, exploring how transformative AI could reshape power, governance, the economy, and individual freedom. It also aired its first television ad during the Super Bowl, aiming to show people how AI can open new possibilities and boost productivity.
Anthropic has made Claude in Chrome generally available on every paid Claude plan, letting Claude act autonomously in the browser instead of requiring approval for every action. The feature is installed from the Chrome Web Store and does not yet run on other Chromium browsers or mobile. A safety classifier validates each action before it is performed, and auto-approval can be switched off in settings to return to manual confirmation.
Indian voice AI startup Ringg raised $10 million from Peak XV Partners as an extension of its Series A, bringing the round to $15.5 million. It processes about 20 million calls a month and is now expanding beyond voice into channels such as chat and WhatsApp, while automating browser-based support for clients like Shell.
Open-source agent tool OpenWorker released a new version that adds multiple security workflow features and can execute tasks on a laptop. Its harness is fully open source, and built-in agents can scan code for vulnerabilities and check dependency poisoning and cloud configuration risks. Users can run open-source models locally, or use a ChatGPT subscription or API key to specify other models.
Starting today, the memory you use in chat is the same as in Claude Cowork: wherever you work with Claude, it starts from what it already knows about you. You can see everything Claude remembers topic by topic and edit or delete any of it. Sensitive subjects such as health or beliefs are not saved to memory by default, with optional sensitive-topic controls.
The LangChain team announced a deeper integration with Airbyte: adding a LangChain destination within Airbyte, complementing the earlier integration that used Airbyte as a LangChain Document Loader. Airbyte contributes hundreds of data sources and scheduled orchestration, while LangChain provides transformation logic such as text splitting and 50+ embedding integrations, helping bring retrieval-based applications from prototype to production.
Anthropic is merging the memory systems used by chat and Claude Cowork, so Claude will always remember what it learned in one area even when used in the other, letting users move seamlessly between research and action without rebriefing. The feature is on by default across Free, Pro, and Max plans on web, desktop, and mobile; iOS and Android users need to update their app. Personal and sensitive information like health data is not saved by default.
OpenAI has introduced the Admin plugin for ChatGPT Work and Codex, letting admins analyze workspace activity, manage members and groups, control access permissions, and adjust usage limits and spending requests within a single conversation. The plugin can automate recurring workflows, such as routing pending usage requests to Slack or Teams for approval, with actions mapped to existing role permissions. To get started, enable the plugin in your ChatGPT workspace settings and install it from the Plugins directory in ChatGPT Work on web or desktop.
At the Hot Chips conference, OpenAI released the first benchmark results for its Jalapeño chip. On SemiAnalysis' InferenceX benchmark, Jalapeño delivered more tokens per user and more throughput per kilowatt than current state-of-the-art inference processors, benchmarked against an Nvidia Blackwell system. Co-developed with Broadcom, it is expected to deploy in very small volumes at the end of 2026, ramping up in 2027.
Presentation startup Gamma acquired Accel-backed design startup Lica to establish its own design research lab, led by Lica's co-founders Priyaa Kalyanaraman and Purvanshi Mehta, exploring fluid, multimodal presentations and other communication formats. Founded in 2023, Lica raised $4 million from Accel, South Park Commons and Village Global.
Apple announced the new Mac Studio featuring the M5 Max and all-new M5 Ultra on August 25, calling it the most powerful Mac ever. The M5 Ultra brings up to a 36-core CPU, an 80-core GPU and 512GB of unified memory. Pre-orders open in 30 countries and regions, with availability from September 22, and the M5 Max model starts at $2,499.
OpenAI outlined an end-to-end AI compute strategy spanning data centers, chips, frontier models, the developer platform, and enterprise products. Its first custom inference chip, Jalapeño, delivered higher peak throughput per kilowatt and lower latency on the InferenceX benchmark with GPT-OSS 120B, while performing strongly on DeepSeek R1 and Kimi K2. The infrastructure portfolio includes Microsoft, NVIDIA, AWS, AMD, Broadcom, and others.
OpenAI says preliminary evidence suggests its upcoming Astra model may meet the Critical cybersecurity capability threshold under its Preparedness Framework, so it now applies the strictest security safeguards to all workloads involving Astra or cyber models. OpenAI strengthened research environment isolation, link monitoring and automated alignment gates, paused reinforcement learning training for two weeks, and kept its largest planned frontier RL run on hold to gather alignment evidence.
Business Insider reports that Hugging Face has been approached about a sale at a valuation of $13 billion or more, though no deal has been reached and the counterparty is unclear. The company was valued at $4.5 billion in its 2023 fundraising round and earlier this year turned down a $500 million investment from Nvidia that would have valued it at $7 billion.
Stanford economists' newly released research finds that generative AI tools are replacing entry-level positions, leaving younger workers facing higher unemployment risk. The study says young employment in AI-affected fields is down 19% compared to more AI-resistant occupations, while older workers so far appear largely unaffected.
OpenAI brings the GPT-5.6 model family (including Sol, Terra, and Luna) to Kiro, its software development agent, saying it delivers more useful work per token and better price-performance. OpenAI and AWS jointly optimized the environment; on Terminal-Bench 2.1, GPT-5.6 Terra completed successful tasks in Kiro at roughly an 82% cost reduction.
Michael Polansky's Outer Biosciences (founded in 2022) sources discarded skin from surgery through vetted biobanks, keeps it alive outside the body for up to a month, and uses it as data to train an AI that predicts cosmetic ingredients. The company has raised roughly $23 million to date, employs 19 people, and four of its six active pipelines look likely to reach commercialization.
Instinct, an AI personal assistant built by a team led by former Sierra scientists (San Francisco-based, operated by Spear Street Technology), entered private access, connecting to users' apps and screens to handle bookings and shopping. Early testers reported issues like data not being deleted, while its terms of service were flagged as granting the company a near-unlimited license to use user materials.
Meta unveiled MetaRoCE, a clean-sheet RDMA transport protocol purpose-built for AI workloads on commodity Ethernet, targeting million-GPU scale without needing PFC or pause frames. In validation on a 64-node AMD GPU cluster, it delivered higher throughput than RoCEv2 and maintained ~86% throughput at 1% packet loss. Meta is releasing the spec and a reference implementation through OCP.
Situational Awareness, the AI-focused hedge fund run by OpenAI alum Leopold Aschenbrenner, is being probed by the SEC. The New York Times reports that the SEC has subpoenaed banks that did business with the fund and warned them to preserve related information. The firm has not been accused of wrongdoing and says it will fully cooperate with regulatory requests.