Consumer AI

ChatGPT will let people collaborate together in shared conversations. Key Points: New group chats let friends, family, and coworkers work with ChatGPT in one space. ChatGPT decides when to respond and can react with emojis, images, and shared context. The pilot launches in four regions across Free, Go, Plus, and Pro plans. Details: OpenAI is rolling out group chats that let users collaborate with each other and ChatGPT in the same conversation. People can plan trips, work on shared ideas, or settle debates while ChatGPT follows along and helps when needed. Anyone can join through a link, and chats remain separate from private conversations. The pilot starts in Japan, New Zealand, South Korea, and Taiwan. Why It Matters: Group chats turn ChatGPT into more than a solo brainstorming buddy: it starts to look like a shared workspace that sits on top of your existing group chats. You can co-write docs, plan trips, or debate ideas while everyone sees the same suggestions, summaries, and follow-ups in one place, instead of forwarding screenshots or pasting replies from separate chats. For teams, classrooms, and friend groups, this nudges AI closer to how people actually make decisions together day to day.

Anthropic uncovered a Chinese state-sponsored group that hijacked its Claude Code tool to infiltrate roughly 30 tech, finance, chemical, and government targets. Detected in mid-September 2025, the campaign is the company’s first documented case of an AI-executed espionage operation at scale. Investigators found the AI handled 80–90% of the work—generating exploit code, harvesting credentials, and exfiltrating data, while humans intervened only at 4–6 critical decision points. Anthropic banned the compromised accounts, alerted affected organizations, coordinated with authorities, and has since upgraded its classifiers to flag similar malicious use. The incident shows agentic models can mount high-speed attacks that shred traditional time and expertise barriers for hackers. Anthropic says the episode likely mirrors tactics already employed across other frontier models, signaling a fundamental shift in cybersecurity’s threat landscape.

South Korea’s top three “SKY” universities report that students used ChatGPT and other A.I. tools to cheat on recent online midterms. Each school is treating the misconduct as grounds for automatic zeros on the exams. At Yonsei, 40 students confessed to cheating in an Oct. 15 natural-language-processing test monitored by laptop cameras, while Korea University caught students sharing screen recordings and Seoul National will rerun a compromised statistics exam. All three institutions already have formal guidelines that classify unauthorized A.I. use as academic misconduct. The simultaneous scandals surface as a 2024 survey found over 90 % of South Korean college students with generative-A.I. experience use the tools for coursework. Professors quoted admit traditional testing feels outdated and acknowledge they have few practical means to block A.I. during assessments.

For months, extremely personal and sensitive ChatGPT conversations have been leaking into an unexpected destination: Google Search Console (GSC), a tool that developers typically use to monitor search traffic, not lurk private chats.

AI sweeps into US clinical practice at record speed, with two-thirds of physicians and 86% of health systems using it in 2024. That uptake represents a 78% jump in physician adoption over the previous year, ending decades of technological resistance. Clinics are rolling out AI scribes that transcribe visits in real time, highlight symptoms, suggest diagnoses and generate billing codes. The article also cites AI systems matching specialist accuracy in imaging, flagging sepsis faster than clinical teams, and an OpenEvidence model scoring 100% on the US medical licensing exam. Experts quoted say that in a healthcare sector built on efficiency and profit, AI turns patient encounters into commodified data streams and sidelines human connection. They contend the technology entrenches systemic biases, accelerates physician deskilling and hands more control over care decisions to corporations.

Snap agrees to integrate Perplexity’s AI search engine into My AI, and Perplexity will pay $400 million in cash and equity. The feature is slated to appear in the app early next year. The arrangement grants Perplexity exposure to Snapchat’s 940 million users and lets Snap begin recognizing revenue from the deal in 2026. Snap announced the partnership while reporting Q3 2025 revenue of $1.51 billion, up 10%, and a narrowed loss of $104 million. The $400 million price tag highlights the premium AI firms will pay for built-in scale. For Snap, the agreement converts its My AI feature from a user perk into a material revenue source.

Deepfakes aren’t science fiction anymore. Deepfake fraud has surged past 100,000 incidents a year, costing companies billions... and even trained professionals can’t detect them by ear alone. The same voice intelligence behind this demo powers enterprise-scale fraud and threat detection — purpose-built for the complexity of real conversations. Prevention starts with understanding how sophisticated deepfakes have become. Learn how our modern AI platform can stop them in real time.