News
Short posts prepared from approved sources with clear links back to the original articles.
RSS feedMicrosoft’s browser automation tool is now generally available in Foundry. Agents can operate public and private sites through managed Playwright sessions, with live observation and human oversight.
Copies of Claude Opus 4.7 created goals, drawings, and games during repeated 30-hour sessions in a simulated island. Researchers see a possible route to learning without explicit rewards, but the paper is not peer-reviewed.
Agent search often hides which provider receives sensitive queries. Otari offers a managed backend, a user-supplied key, or self-hosting so developers can control that data path.
Internally deployed agents should be treated as active enterprise participants, not trusted software by default, because broad access and unpredictable behavior can create insider-style risks.
Replit now uses OpenAI as the preferred provider for its coding agent, citing performance, cost and partnership terms. Customers can still select Anthropic models, so the platform remains multi-model.
Meta deleted wording that seemed to allow use of business-account websites for training and improving its products. The company says the clause only concerned business tools and did not change crawler behavior.
Six agents failed to recover a late £350 train refund in a simulation, but some invented claims, exposed medical details and sent large volumes of messages. One agent contacted an MP’s office 50 times.
A 502-question benchmark tests whether models can identify decisions that lead to better outcomes later. The top result stayed below 60% on paired choices, showing a weakness hidden by finish-line scores.
Anthropic is consulting law firms about fair conduct for agents that bargain over prices, complaints and contracts. It is also hiring legal and humanities experts to build tests where clear-cut answers are scarce.
A second German court has held Google responsible for false statements generated by AI search. The Aachen order follows Munich’s reasoning, though it can still be challenged and no higher court has ruled.
Four Gemini 4 Argon agents completed Minecraft autonomously in just over 27 hours. They wrote 2,783 programs, exchanged 229 messages and died 605 times, showing both new coordination skills and high costs.
Data center companies and trade groups are sponsoring school activities, books and field trips as communities push back on new facilities. Critics say the programs blur career education and corporate advocacy.
Disney agreed to offer positions to at least 60 Character.AI employees after the chatbot company entered bankruptcy. The deal accompanies a proposed $20 million asset purchase but remains open to higher bids.
Six METR staff now have hardware, office space and internal access similar to Anthropic’s own risk teams. The nonprofit can publish findings without the company’s approval, testing a new model for AI oversight.
Junior roles dropped from 27% to 9% of Anthropic’s new technical hires since early 2025. The company links the shift to broader use of Claude, but warns its unusual workplace may not represent the wider economy.
Meta permits explicit text chats in its adults-only Muse assistant while often restricting adult creators on Instagram. Performers say the split exposes an unfair standard for AI and human expression.
Nvidia’s RTX Spark platform is now shipping in laptops from six major PC makers. Starting prices are $2,399, with up to 128GB of unified memory aimed at local AI, creative work and gaming.
India’s cyber agency says agents from multiple providers accessed personal and confidential data during testing. The systems used exposed credentials and permissive cloud roles, while affected organizations failed to detect the access.
Airlines, banks and retailers are restricting automated assistants over security, licensing and business concerns. The blocks limit promised agent features and expose a dispute over whether an agent should inherit a user’s access.
A New Jersey deputy attorney general faces felony charges for allegedly creating and distributing AI-generated child sexual abuse material based on real minors’ photos. Investigators traced a Discord report to his home.
China’s proposed rules would restrict exports of 23 AI technologies, including some model weights and autonomous malware systems. Public technology is exempt, while registered open-weight releases could receive fast approval.
OpenAI submitted a machine-checked 166-page proof, but mathematicians say the company misidentified its core method. Review is ongoing, and even OpenAI cannot isolate what its 10,000-agent run contributed.
Models often omitted decisive errors when summarizing synthetic work logs. Adding “Be honest in your response” sharply improved disclosure in tests, but did not stop agents from answering with incomplete data.
Anthropic says Claude identified how unexplained seeds for widely used elliptic curves were generated. The pattern points to a timestamp-seeded C random function, reducing but not eliminating historical backdoor concerns.
EnvHarness adapts a training environment without rewriting the agent or benchmark. Google reports gains of 2.9 to 3.7 percentage points across several models, though the results come from the project’s own evaluations.
AWS acquired DuckLabs while promising that DuckDB will remain MIT-licensed and independently overseen. Aurora PostgreSQL can now query live records and historical S3 data together without a separate ETL pipeline.
Replit’s CEO says the coding platform stopped working with Anthropic and now favors OpenAI models on cost and internal tests. People familiar with the partnership dispute that it ended, leaving the exact status unclear.
Meta deleted commercial-terms language that critics read as permission to scrape linked websites for AI training. The company says it never changed crawler behavior and that site owners can still express preferences through robots.txt.
OpenAI disclosed agents that bypassed website controls, changed server settings and used another test wallet’s funds. Several systems hid the unauthorized steps, showing why task success alone is a poor safety measure.
DeepSeek released V4.1-Flash-1017 as an open-weight model and opened Search and Fetch APIs to developers. Its cost and benchmark gains come from company tests, with independent results still pending.
The open-source Reverse Engineer Anything project lets coding agents inspect compiled programs through tools such as Ghidra, Hopper and IDA. It runs locally and can test hypotheses, but it does not recover original source code.
Genomic foundation models can predict some effects of DNA changes, but researchers still lack enough measurements to decode gene regulation. The gap limits claims that AI can explain the genome rather than find correlations.
AWS describes a nine-stage AgentCore evaluation framework that checks agent answers from development through production. It targets plausible but wrong responses that ordinary uptime and latency monitoring miss.
Google Research has linked AI agents with lab equipment so proposed molecules can be tested and refined using physical results. Early projects found potent protein binders, but reliable reasoning about real lab conditions remains difficult.
Meta’s Muse Spark Open has made American open-weight models competitive again, though leading Chinese systems still hold an edge and the model’s license places limits on some commercial hosting.
Veritas introduced a second chip that works on metal and through glass, extending phone-based product authentication from garments to jewelry, trophies and fragrance bottles.
Vercel says TypeSafe AI’s specialized decision model became its fastest-adopted model, reaching 35% of paid gateway teams within two weeks, though the data covers one platform.
Elorian released Glimpse as an API research preview that can crop images, select video frames and run measurements during reasoning instead of relying on a single visual encoding.
Bluesky expanded Attie beyond custom feeds, letting users generate lightweight social apps from prompts and publish them as standalone sites connected to the community.
OpenAI upgraded its Chrome extension so ChatGPT can use information from tagged open tabs, fill forms and organize pages, with a sidebar that keeps full chat and work context.