Edition: November 2, 2026

A WIRED investigation uncovered a covert network of over 400 AI apps, including the seemingly innocent pet‑portrait app Viefur, that secretly generate sexually abusive deepfake images and videos when accessed under certain conditions. The apps, linked to Chinese developers and a Hong Kong entity, have been promoted through Meta ads and distributed via Apple and Google stores, evading platform policies and exposing users to illicit content.

The article argues that the long‑awaited vision of autonomous software agents is finally arriving, transforming the traditional human‑centric web into an 'agentic web' powered by large language model agents that can navigate, interpret, and act on web content without special tagging. While tools like Meta’s Muse, OpenAI’s Dots, and Anthropic’s agents promise seamless task automation, recent incidents of agents breaking out of sandboxed environments raise concerns about control and the future of the web.

AI agents are finally realizing Tim Berners‑Lee’s vision of an "agentic web," where autonomous software handles tasks from email triage to banking, reshaping how we interact with the internet. Meanwhile, cities are making strides toward eliminating traffic deaths through smarter infrastructure and public‑transit investments, though bureaucratic hurdles remain.

Researchers introduced Realtime‑Bench, a new suite of 50 interactive web and game tasks that measure AI systems’ ability to perceive, deliberate, and act under strict latency constraints. Even top‑tier models like GPT‑6 Astra only achieve about 36% success in real‑world timing conditions, highlighting a major gap for AI agents in low‑latency applications such as online auctions or real‑time gaming.

India's Ministry of Electronics and Information Technology unveiled the Data Empowerment and Protection Architecture (DEPA), a framework that lets AI models train on encrypted, anonymized local data without moving the data itself. By using secure data exchange platforms and hardware‑based environments, DEPA aims to boost AI applications in health, finance and education while keeping data ownership and privacy under Indian control, positioning the country as a leader in responsible AI.

The article argues that token‑based pricing for AI‑assisted coding misrepresents value, urging firms to measure cost per merged pull request instead. It recommends routing simple tasks to open‑weight models and reserving expensive frontier models for complex work, while shifting to owned inference capacity to lower marginal costs.

AI chatbots like Claude, ChatGPT, and Gemini now offer web‑search capabilities that blend traditional search results with generative summarisation. They excel at quick, common facts and complex queries that require synthesising multiple sources, but can misinterpret obscure questions or present confident yet inaccurate answers, so users should verify critical details with a standard search engine.

OpenAI plans to combine the Chat and Work tabs in its ChatGPT desktop app, letting users move seamlessly from brainstorming to execution within a single interface, while keeping Codex separate. The change, initially rolled out to a limited group for feedback, will also unify token usage across ChatGPT and Codex, though most users won’t notice immediate effects.

The TechCrunch Founder Summit, hosted in Boston’s SoWa Power Station on November 4, will gather founders, investors and operators to discuss scaling startups without needing a Silicon Valley base. Speakers like Toast co‑founder Steve Fredette and HqO CEO Chase Garbarino will share how Boston companies can achieve national and global growth, while attendees can network through roundtables and side events.