
What happened
Forever Security's Gal Weizman detailed EMBRACE, where embedded browsers in Cursor, GitHub Copilot, OpenAI's Codex, Microsoft's VS Code and Anthropic's Claude Desktop loosen real-browser safety rules; Codex (reported July 27) and VS Code (reported September 10) remain unpatched.
Why it matters
Until fixes ship, developers using these five tools may be exposed to attacks that a normal browser would block, according to the researchers.
WHO IT HITSDevelopers using Cursor, GitHub Copilot, OpenAI's Codex, Microsoft's VS Code or Anthropic's Claude Desktop could be exposed to attacks that a normal browser would block until patches ship.
Summaries like this, every morning.
The EMBRACE findings arrive amid a run of agent-security disclosures, including Anthropic's own chained-exploit write-up a day earlier. Forever Security's report details how each vendor loosened one of three browser safety rules so the AI could preview a project or read what a tab shows. Cursor and GitHub Copilot have since been patched, but Codex and VS Code remained unpatched at publication, with VS Code at around 40 million developers and Codex at more than 5 million weekly users. The inconsistent responses—Microsoft and Anthropic rejecting reports as workspace trust issues, OpenAI silent after the July 27 disclosure—suggest that triage practices, not just code, are part of the story.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Cognition released Devin Security 2.0 on October 26, 2026, adding deep scans powered by frontier cyber models…

SB Intuitions announced Sarashina3.2 mini thinking, the Sarashina series' first thinking model, available via…

A Reddit poster traced failed agent runs and concluded that pasting bad code under a "DON'T do this" comment l…

The guide, based on an October 12, 2026 session, says to press Shift+Tab once so the display changes from "aut…

AI Labo's E05 verification series set up a fully offline sandbox VM and tested two agent design patterns — rol…

A Zenn article laid out a sealed-environment AI setup — one gateway into Claude Code via Amazon Bedrock, with…
