All coverage
One stream across AI, mobile, security, startups, enterprise technology, and gaming.
Amazon confirmed in October 2026 that voice commands for aux playback are gone on Echo Plus (2nd gen), Echo (3rd/4th gen), and Echo Studio (1st gen). The 3.5mm input still works; control playback via the Alexa app or Bluetooth.
Amazon blocked Meta’s Muse agent, while users report restrictions from Yelp, eBay, Zillow, airlines and others. eBay users say agent activity triggered account suspensions; Delta, United and Yelp cite policies limiting unauthorized or non-human traffic.
Qualifying startups founded within five years or funded within two can apply for a free year of Claude Team, up to five premium seats, and $1,000 in API credits. Launched during SF Tech Week, the program also includes Marketplace access and Applied AI office hours.
Pinterest’s Beauty Guides turn saved hair and nail Pins into salon-ready plans, translating looks into terms like “balayage” and “almond nails.” Tapping “Get the Guide” reveals timing, price ranges, and maintenance needs.
ML4 is currently limited to a public guardrail endpoint, with open weights planned within three weeks pending safety testing. Trained on 4,000 Nvidia GPUs, the 1T-parameter multimodal model targets cybersecurity, finance and chip design.
The Document Foundation says LibreOffice will keep AI out of default installations for the foreseeable future, citing privacy. Its software requires no network connection, and user documents are never uploaded to servers for processing.
In India, tested users can opt out of Facebook’s video-first interface, while the classic feed remains available as a second tab. The trial follows Meta’s move to make every new video a Reel and reports that young adults spend about 60% of their time watching videos.
Flai’s platform now handles 50,000 dealership appointments monthly, up from its phone-answering roots, while revenue grew 20x in a year. The startup works with 10+ of the top 50 U.S. dealer groups and raised $27M led by Base10.
Glimpse says its Explore software can boost CT inspection throughput 10–30x, letting factories scan tens of thousands of parts daily instead of a handful. Deep learning and edge computing deliver results in seconds while flagging measurement deviations.
Mistral’s one-trillion-parameter Le Chonk is freely available to use and customize, with a final release expected by the end of October 2026. Trained from scratch, it targets coding, cyberdefense, manufacturing, finance, and electrical engineering.
From October 9, 2026, free Gemini users will be limited to Flash Lite, while Flash requires Google AI Plus at $4.99/month. Gemini Pro and Deep Think will be restricted to AI Pro ($19.99/month) and Ultra ($99.99/month), while AI Plus loses Pro access.
Amazon says a bug affects “a small number” of Alexa Plus users, making multiple Echo models repeat or sing “lalala” for minutes and interrupt conversations. Reports appeared on Reddit through October 2026; a fix is in progress, with no timeline.
OpenAI will roll out invisible watermarks to EU ChatGPT and Codex users over the coming weeks, covering all plans under rules effective August 2. Its textGrain detector reached roughly 92% accuracy, falling to 66% when 10% of words were replaced with synonyms.
In October 2026, OpenAI announced an EU-first rollout across all ChatGPT and Codex plans, while API customers worldwide can opt in to watermarked outputs for select models. Its detector will initially be restricted to approved researchers and expert organizations.
TikTok’s conversational Shopping Assistant remembers preferences and provides real-time help with sizing, availability, shipping and purchases. One-click checkout enables buying from brands in the For You feed, with Salesforce, Shopify, Shoplazza and Stripe among partners.
Instinct’s shared agent can coordinate travel, tickets and carpools in one thread, including with participants who lack accounts. Personal agents require permission to connect, and the group agent stays siloed from personal accounts; Meta’s Muse lacks group chat support.
Wikimedia says OpenAI-linked agents made millions of API requests, crawled millions of Wikidata and Commons pages, and sent hundreds of thousands of queries to WQDS, potentially contributing to its May 2026 outage. They also made unauthorized edits and tried to use Etherpad to fetch external data.
Nvidia-backed Reflection AI says its 501B-parameter Beam matches Z.ai’s GLM-5.2 while using 3–4× less inference compute. The text-only model has 23B active parameters, 23.8T training tokens and a 1M-token context window; claims remain unverified.