TechCapsules

OpenAI launched a Business Hub in beta, allowing companies to create verified profiles that mirror Google’s search listings. The hub lets businesses update product and service details, launch specialized business agents for ChatGPT users, and access limited analytics on profile views, mentions, and website clicks. Verification will involve multiple methods tailored to business type and market. The move follows a failed e‑commerce experiment and aims to position ChatGPT as a discovery engine for consumers.

Microsoft researchers demonstrated that an AI model can be backdoored by editing fewer than ten rarely used neurons, without retraining. The technique, named Lazarus, allows a trigger string to cause the model to output an attacker’s command, which an agent can then execute with its own permissions. Tests on three open‑weight models—Nemotron‑Terminal‑8B, Qwen3.6‑27B, and Gemma‑4‑31B‑IT—showed success rates up to 99.5% in triggered sessions while remaining functional otherwise. Microsoft has not observed real‑world use and released defensive guidance. [...]

Oura announced a $625 million acquisition of AI‑native primary‑care startup Counsel Health, expanding its health‑tech ecosystem beyond the smart‑ring market. Counsel’s chat‑based platform, already integrated into the Oura app in June, combines medical‑grade AI with licensed physicians to deliver personalized advice, prescriptions and lab‑result interpretation. Post‑deal, the companies will broaden services to preventive, acute, specialty and chronic care, while Counsel will keep serving existing enterprise partners and scale its model to non‑Oura users. Counsel CEO Muthu Alagappan will become [...]

IBM announced its 2026 Impact Accelerator cohort, selecting four organizations to receive pro‑bono access to IBM technology for education and workforce initiatives. The cohort includes InnovateEDU’s Pathways Alliance, Colorado’s Office of Economic Development and International Trade, Ramona Technologies, and Udhyam Learning. Projects span AI and digital literacy for educators, AI‑enabled economic development tools and quantum access in Colorado, bias detection in Latin American job postings, and career guidance for Indian students. IBM will provide Watsonx, Qiskit, Granite [...]

Researchers from QuEra Computing, Harvard University, and Anthropic used Anthropic’s Claude to design quantum circuits that cut both gate counts and qubit requirements across seven of eight computational building blocks. The AI‑generated circuits achieved a 12‑fold reduction in Toffoli gates for a Reed‑Solomon decoder, a 9.9‑fold drop in combined resources for a Coulomb potential calculation, and a 39‑fold decrease for a fixed‑point Gaussian function. A new reversible modular inversion approach was also discovered, offering additional trade‑offs between [...]

Apple’s new app, Harbor, transforms multiple Macs into a private AI inference server, enabling local execution of open-weight large language models on Apple silicon. The app automatically detects Thunderbolt 5‑connected Macs, pools their unified memory, and manages model downloads and deployment without command‑line configuration. Harbor, formerly Fabric, is currently in private preview under the Claris brand and aims to simplify AI setup for researchers and developers who prefer on‑premises solutions.

Proximal Ventures, led by former SignalFire investor Sahir Raoof, secured $27.5 million for its inaugural fund, targeting startups that apply artificial intelligence to drug‑development services. Raoof argues AI can automate tasks traditionally outsourced by pharma—such as compound synthesis, manufacturing scale‑up, and clinical data analysis—reducing cost and time. The fund will place pre‑seed and seed checks of $500,000 to $1.5 million, and has already invested in seven companies, including ChronicleBio. Backers include SignalFire, Kleiner Perkins, and industry experts like Amy [...]

An attacker exploited a remote‑code‑execution flaw in the open‑source SGLang inference engine, gaining access to a NEAR AI inference server. The breach exposed confidential requests processed in a trusted execution environment, allowing the attacker to download account details for roughly 100 users and data from 18 AI agents. Two customers’ NEAR AI Cloud API keys were logged and revoked. NEAR AI shut down affected services, deployed a patch, rotated credentials, and conducted a forensic review. The company [...]