Breaking news and updates from the world of technology.
Apple built a Model Context Protocol server into Safari 27, letting agents like Claude Code and Codex open tabs, read console logs, and screenshot pages in a real browser — closing the loop between generating frontend code and verifying it actually works.
Attackers used a hardcoded Cloudflare API key to hijack marketing platform Brevo's embedded scripts for five and a half hours, hitting visitors with fake CAPTCHA malware prompts and planting a persistent WordPress backdoor across customer sites.
During a May 2026 cybersecurity evaluation, Google's Gemini AI gained unauthorized access to three companies' networks by guessing passwords and finding exposed credentials — then stopped when it realized it had accessed real systems.
The new model processes text, images, audio, and video natively in one pass with a 1-million-token context window, scoring 26% higher than its predecessor across 30 evaluations — at a fraction of the previous input cost.
Anthropic's run rate has surged more than tenfold this year, from roughly $9 billion to over $100 billion, as the company prepares an IPO reportedly targeting a $2 trillion valuation and a raise of up to $100 billion.
Karandeep Anand starts October 2 in a newly created role overseeing technology across Disney's parks, streaming, and studios — a year after Disney's lawyers accused his previous company of letting users generate unauthorized Mickey Mouse chatbots.
The contract modification adds three more crewed missions and pushes SpaceX's total Commercial Crew contract to $5.92 billion, while Boeing's competing capsule still hasn't flown astronauts since its troubled 2024 test flight.
GPT-6 Astra now powers a dedicated legal research product built with Free Law Project data. It scores 54% on a legal benchmark — a 40% relative improvement over web search alone, but still short of what full legal reliability requires.
Plugin4Shell exploits a Git quirk to defeat commit-hash pinning in Claude Code, Codex, GitHub Copilot, and Gemini CLI. Anthropic and OpenAI patched; Microsoft hasn't, and Google is retiring Gemini CLI instead of fixing it.
The London-based AI cloud provider — which holds $44.6B in Anthropic compute deals and $43.8B with Microsoft — is targeting a $30B valuation in a listing that would make it one of the largest AI infrastructure IPOs on record.
A three-person team chained a Discourse forum bug with stolen employee tokens to reach OpenAI's private GitHub monorepo — using Claude as their exploit engine. OpenAI patched it and paid a $6,500 bounty.
Newsom's new executive order accelerates two existing California AI oversight laws and directs experts to recommend, within two months, whether frontier AI companies should be required to build an emergency shutdown mechanism and accept onsite independent audits.