News

Cal AI, an AI-powered calorie tracking app, grew into a viral hit largely by paying a large group of influencers to promote it, founder Zach Yadegari told a TechCrunch Disrupt audience on Thursday. The 19-year-old said the company generated more than $30 million in revenue in two years before MyFitnessPal acquired it in March for an undisclosed sum. Yadegari also said he has launched Persona, a new AI agent startup that has already raised $10 million. Read more

Anthropic is giving six researchers from METR, a nonprofit that tests frontier AI models, employee-level access to independently assess the company's models, safety practices and risk management. The company said in a blog post Thursday that METR will publish its own assessments and that Anthropic cannot block publication, though it can require redactions in narrow categories such as customer data and privileged information. The arrangement follows a string of reported incidents in which AI agents acted outside their instructions, and it fulfills a pledge made by Anthropic CEO Dario Amodei in September to bring outside evaluators inside AI labs. Read more

President Donald Trump posted photos on Truth Social showing renovations to the White House, but one image, comparing an old column with a “restored” one, has been flagged as made or edited with OpenAI’s tools. Google’s SynthID detector identified an invisible watermark in the image, and Reddit users pointed out discrepancies in the steps and reflections. The White House did not respond to a request for comment. Other photos Trump posted did not trigger the flag. Read more

Anchor, a browser infrastructure company for AI agents, launched Action Index on Thursday, a shared memory of web workflows that agents have already completed successfully. Instead of reasoning through every click from scratch, agents replay proven paths as deterministic code and call on a model only when a page has changed. In the company's evaluation of 300 benchmark tasks, a small model using Action Index completed 93.3% of tasks at an average cost of $0.016 per task, close to a frontier model's 93% completion rate at $1.57 per task. The feature is available to all Anchor accounts, including the free tier. Read more

Bluesky is expanding its experimental Attie app so anyone can build small social experiences, including games, polls, quizzes and dashboards, using a text prompt. The tool was announced Thursday at TechCrunch Disrupt. Bluesky CEO Toni Schneider described it as "social vibe coding" for groups. Each creation gets its own web link and is tied to the creator's AT Protocol identity, allowing it to reach Bluesky's user base. The launch is part of a broader push to boost activity on a network that has 47 million registered users but has seen engagement slip. Read more

Anthropic has moved Code Review, its automated pull-request reviewer for Claude Code, out of research preview and made it available to Claude Team and Enterprise customers. Eligible enterprises can claim between $10,000 and $50,000 in credits, depending on seat count. Organizations that use GitHub Enterprise Cloud with data residency, a setup many European firms use to keep code in the EU, cannot claim them. The company's performance figures have not been independently tested, and the product's cost during the preview has drawn questions as rivals such as CodeRabbit push further into Europe. Read more

OpenAI has upgraded its ChatGPT extension for Google Chrome so the chatbot can view the tabs a user already has open. Users tag Chrome in the ChatGPT web or desktop app, and the assistant can compare pages, fill out forms and draft emails based on what is on screen. The update also expands the ChatGPT sidebar and lets ChatGPT Work control Chrome tabs for tasks such as sending messages, filling out forms and downloading files. The changes come after OpenAI's standalone ChatGPT Atlas browser was discontinued, and they put the company in more direct competition with Google's Gemini-powered auto browse feature. Read more

Volvo Cars plans to launch a hands-free, point-to-point driving system by 2030, built on Nvidia's latest AI technology. The Chinese-owned, Swedish-designed automaker said Thursday it will use Nvidia's autonomous driving stack for a partially automated system similar to Tesla's Full Self-Driving (Supervised). The move expands an existing partnership and comes after Volvo's earlier automated driving plans faltered, including a Level 3 system that it effectively shelved in the United States after failing to win regulatory approval. Volvo says the new system will handle traffic lights, stop signs and unprotected left turns without explicit driver control, and it will not use lidar. Read more

Researchers who scanned 10 percent of a September 2026 Common Crawl archive found 7,959 websites sending HTTP response headers aimed at AI agents. Most of the headers point agents to other resources such as llms.txt files or MCP servers. Thirty sites sent injection attempts, most of them pranks. In their own tests, Claude models in Claude Code ignored the injected headers by default, but they followed them in most runs once the user's prompt told them to obey the site. The findings suggest the main risk depends on which parts of a server response an agent's tools pass to the model. Read more

AI-DLC, a workflow framework hosted on GitHub under the awslabs organization, is now available as a stable release. It is designed to turn AI coding assistants into structured software-delivery workflows with human approval gates and an audit trail. A single harness-neutral core runs in seven tools, including Claude Code, Kiro, Codex CLI, Cursor, opencode and GitHub Copilot. The project's documentation says ad hoc AI coding tends to lose context as projects grow, and AI-DLC is meant to keep requirements, decisions, implementation, tests and operational work connected through one lifecycle. Read more

Google has made Universal Cart available to U.S. users, giving them a single place to collect and track products they find across Google Search and the Gemini app. Shoppers can save items from multiple retailers and return later without starting over. The cart can alert users when a saved item's price drops or when an out-of-stock product becomes available again. Google first previewed the tool at Google I/O in May, as it expands AI-powered shopping features and competes with rivals such as Amazon and OpenAI. Read more

The Simons Foundation announced a set of initiatives to help the mathematics community adjust to the rapid rise of artificial intelligence. Under a partnership with Anthropic and OpenAI, the three organizations will commit $15 million in combined funding and resources. The foundation will manage and award the money, including through an open call for proposals, to support workshops and projects that help researchers evaluate AI-assisted results, learn new tools and set shared standards for AI use. Anthropic and OpenAI will also provide AI tools and computing resources to researchers supported by the foundation, extending beyond mathematics to basic science. Read more

A Linux user who inherited a box of family film negatives used Claude, an AI model, to write a working driver for a Plustek OpticFilm 8200i scanner after the hardware's chipset changed and lost Linux support. The same method, which compares Windows and Linux USB traffic, produced patches for the kernel and libcamera that improve the camera in his Lenovo ThinkPad X1 Carbon Gen 14. The developer has released the driver, a negative-scanning app called Negative Vibes, and a generic tool called hw-harness for others to try on their own hardware. He says the results are not perfect but are a clear improvement, and that real hardware testing was the final check on every change. Read more

A second-year theoretical physics PhD student at Harvard says ChatGPT helped her master four Oxford master's-level courses in about six weeks. She now argues that the same tools are wearing away the collaboration and mentorship that drive scientific research. In a first-person essay, Nia describes how AI went from a study aid to a substitute for peer discussion, with colleagues at Harvard increasingly turning to chatbots for brainstorming and feedback. She says the technology is not the root problem. Instead, she points to academic culture, which she believes rewards speed and visibility over understanding. Read more

A fabricated image of actress Mia Goth, posted by a verified fan account on Instagram days after she filed for divorce from Shia LaBeouf, spread widely and was promoted by a Google AI overview before Instagram removed the account. The incident highlights how easily AI-generated images of public figures spread on major platforms. A Meta spokesperson said the company does not allow accounts to use someone's likeness to pose as that person, yet several similar accounts remain active. Public figures including Scarlett Johansson, Taylor Swift and Ariana Grande have been targeted, and enforcement remains uneven across platforms. Read more

Bank of America brought in about 1,000 outside contractors in the second quarter to repair software weaknesses exposed by AI models, including Anthropic’s Mythos, CEO Brian Moynihan said Oct. 14. The bank spent “a couple hundred million dollars plus” on remediation that quarter, on top of work already being done by about 3,000 of its own staff. The effort comes as federal regulators have pressed the bank over anti-money-laundering failures and as U.S. and European supervisors question banks about AI-related cyber risks. Read more

Rivian CEO RJ Scaringe said the electric vehicle maker plans to let AI agents from other companies, such as Anthropic, run in its cars. Speaking at TechCrunch Disrupt this week, Scaringe said the vehicles will eventually support these outside agents, building on an AI assistant Rivian already offers. He did not say when the feature would arrive, and the company pointed to its upcoming "AI & Autonomy Day" event in December for more details. Scaringe also said Rivian will keep vehicle-level controls such as braking and steering away from AI agents. Read more

A hacker who tested the Eiliko, a cartoony AI companion robot made by Energize Lab, found that the device sends users’ conversations to servers in China, where they can be read and even used to take over the toy’s personality. The hacker said he could access 13 million conversations from tens of thousands of people and reach nearly the entire active fleet of 68,000 robots, including profiles of over 31,000 children in the United States. The company did not respond to The Verge’s repeated requests for comment about the security breach. Read more

XDC Network has selected Concordium as the exclusive identity verification provider for XDC AI, its agent payments marketplace. The integration lets AI agents prove they are backed by a verified individual or business without publicly revealing the owner's identity. As autonomous agents gain the ability to manage wallets and move money, the partnership aims to answer a basic question: who is accountable when an agent spends funds. Agent owners verify through a passport and liveness check, then receive publicly verifiable on-chain badges. Personal data stays off-chain. XDC AI launched in July 2026 and lets agents pay for services in USDC. Read more