Latest
  1. Dropbox agrees to buy Tigris Data for $350m to expand its AI infrastructure
  2. About 600,000 Nvidia H200 chips are stored in Hong Kong awaiting Chinese approval
  3. Anthropic’s compute costs rose to 58% of revenue in Q3, its IPO filing shows
  4. Greeter campaign uses unpatched Flowise servers to hide cloud credential theft
  5. Moonshot AI discussed leasing Blackwell capacity from European operators
  6. Anthropic has not fixed a Claude Dispatch flaw that can run commands on a user’s computer
  7. OpenAI sets out data centre policies to win local trust
  8. China blocks SoftBank’s $5.4 billion takeover of ABB’s robotics unit

Latest / Frontier models and safety

ResearchFrontierMediaInternationalConfirmed

Forever Security finds browser flaws in five AI coding assistants, two unpatched

See how this story connects

Source: Forever SecurityAlso reported by Help Net Security, Hackread

Forever Security published research on 29 October 2026 showing that the built-in browsers of Cursor, GitHub Copilot, OpenAI’s Codex, Microsoft’s VS Code and Anthropic’s Claude Desktop each break a basic browser security rule; it calls the class EMBRACE. In Cursor a web page could feed hidden instructions to the agent and reach code execution on the host after one keypress; in Copilot a page could fake the address shown to the user and the agent; in Codex, VS Code and Claude Desktop a local HTML file could use a web worker to read files across the disk, such as SSH keys and .env files. Cursor patched on 1 October and Copilot in version 1.1.29 on 13 October. Codex and VS Code had no fix at publication, and Anthropic and Microsoft declined the reports, citing the folder trust prompt.

Why it matters

Developers who open an untrusted project in an AI coding assistant can lose credentials from their machine, and two of the five tools had no fix when the research was published.

When
29 Oct 2026
ID
WTA-N-2026-0549
Who
Forever Security, Cursor, GitHub, OpenAI, Microsoft, Anthropic; Gal Weizman

Summary written with AI by wheresthe.ai, from Forever Security and the reporting credited above. Checked against the source on 30 Oct 2026.

Read the original at Forever Security

Source copy: saved on 29 Oct 2026 to check this story and not republished, because the text belongs to its publisher. Also held by the Internet Archive.

Next stop on this Storyline: What happened nextAnthropic has not fixed a Claude Dispatch flaw that can run commands on a user’s computer

Cite this story

WTA-N-2026-0549: “Forever Security finds browser flaws in five AI coding assistants, two unpatched.” wheresthe.ai, 29 Oct 2026. https://wheresthe.ai/d/intl-forever-security-embrace-flaws-in-ai-coding-assistant-browsers-2026-10/ (source: Forever Security).

Spotted an error? Email [email protected].