All coverage
One stream across AI, mobile, security, startups, enterprise technology, and gaming.
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.
Menlo Ventures invested in Factory’s latest funding round at a $5 billion valuation, days after Vinod Khosla called the AI coding startup a “struggling second tier competitor.” The amount wasn’t disclosed, but Menlo said it was significant.
After roughly six months of beta testing with Snowflake, Snorkel and Capgemini, the tool conducted more than 500,000 interviews. It combines a real-world coding task, AI assistant and follow-up questions to assess reasoning and “AI fluency,” not just output.
On Politico’s Decoded podcast, Sam Altman said society should tolerate some AI-driven hacks and scams because people will do “orders of magnitude” more good. He called OpenAI “pragmatic centrists” and warned overregulation could concentrate power.
During a Vanity Fair interview in October 2026, an OpenAI publicist twice interrupted editor Mark Guiducci as he raised a ChatGPT-linked suicide case involving Laura Reiley’s daughter. Sam Altman said mental-health policy questions are among the hardest and user data should be shared only with consent.
Bootstrapped by New York sisters Baila and Stanford-trained computer scientist Sumrin Mudgil, the iOS and web app uses behavioral science and Socratic prompts instead of companion-style emotional dependency. Safety guardrails can direct users to 988.
Attackers exploited a Danish company’s lawful CPR access to steal records on about 8 million people, including names, addresses and social security numbers. Discovered October 2, the September breach is believed to be Denmark’s largest.
Lola Vision’s compiler toolchain targets the roughly 200 hours engineers may spend manually configuring AI models for new hardware, translating customer code for specific chips. Founded in 2024 by Tayo Adesanya, the Washington, D.C., startup is focusing on aerospace and other mission-critical industries.
The test begins in late October 2026 in the U.S., placing clearly labeled ads beside AI-generated images without affecting ChatGPT’s responses. OpenAI says the format could eventually reach its 1.2 billion weekly users worldwide.
OpenAI will test image-based sponsored ads in ChatGPT in the US later in October 2026, initially alongside image generation. Ads will stay separate from generated images and won’t affect answers; Plus, Pro, and Enterprise users won’t see them.
On October 5, Google acknowledged that some complex Android apps and games may perform poorly on Intel-based Googlebooks because they were designed for Arm chips. Of five launch models, three use Intel processors and two use Qualcomm Snapdragon.
Safeworld runs robots’ actual software through thousands of simulated scenarios featuring realistic human behavior, from blind factory corners to falls. Founded by Carnegie Mellon’s Ding Zhao, it raised more than $12 million from Shine Capital and a16z Speedrun.
Verifact Markets, launched in October 2026, lets traders submit evidence for “true” or “false” positions on disputed past events, with rewards planned for documentation that helps resolve markets. In beta, Covid-19’s lab-origin claim traded at 76¢ “true” versus 32¢ “false.”
Google suspended its Open Source Software Vulnerability Rewards Program on October 1 after engineers and maintainers were overwhelmed by a surge of automated reports that were mostly invalid or hallucinated. An update is promised in Q1 2027.
Announced October 4, 2026, the force puts National Intelligence Director Jay Clayton in charge, with FTC Chair Andrew Ferguson, Emil Michael and Scott Kupor as vice chairs. It has 120 days to report on AI’s risks and opportunities.