Forever Security · Oct 30, 2026 · 5:17 AM

EMBRACE research breaks the built-in browsers of Cursor, Copilot, Codex, VS Code and Claude Desktop

Published Oct 29, the research shows a booby-trapped page got Cursor's agent to run a host command (patched Oct 1), and a local HTML file in Codex, VS Code or Claude Desktop could read and exfiltrate files across the disk, with Copilot spoofing the address bar (fixed in 1.1.29). Codex and VS Code were still unpatched at publication, so be careful opening untrusted HTML in their built-in browsers (Help Net Security).

Read at forever.security

More news

View all

Kimi Code 2.4.0 imports your recent Claude Code and Codex sessions so you can keep working in Kimi

Released Oct 28, kimi import sessions (or /import sessions ) converts up to 100 Claude Code and Codex sessions from the last 15 or 30 days into regular Kimi sessions in their original project folders, locally and without calling a model or touching the originals. The release also makes Plan mode user only, so the agent can't enter it or auto approve plans on its own (changelog)…

Kimi Code Docs (Moonshot AI) 19 minutes ago

Codex CLI 0.166.0 adds keyboard transcript search and keeps approvals when you resume or fork

Released Oct 29, the TUI now lets you select and search the transcript by word, line or page with configurable shortcuts, and adds codex queue resume and codex doctor network only . Resumed or forked threads now keep their approval settings and reasoning token counts, so they don't compact early…

GitHub (openai/codex) 2 hours ago

Claude Code v2.1.314 adds a setting that blocks file writes outside your working directories

Released Oct 30, the new permissions.blockWritesOutsideWorkingDirectories makes Write, Edit and NotebookEdit refuse paths outside the working directories in every permission mode, and VS Code and Claude Desktop's Code tab now summarize idle 200K+ token conversations so your next message doesn't pay for an expired cache. It also fixes claude p silently exiting 0 when startup fails…

GitHub (anthropics/claude-code) 3 hours ago

OpenCode v2.2.1 adds /goal, which keeps a session working until a reviewer says the objective is met

Released Oct 29, /goal <objective in the TUI or Desktop has the session's own model review progress after each turn and keep going until it's done, blocked or out of rounds, with /goal pause , /goal resume and /goal clear to steer it. The release also adds opt in scheduled prompts via /schedule and an experimental image generate tool…

GitHub (anomalyco/opencode) 4 hours ago

&Truth exits stealth with $70M seed to mathematically prove AI-written code is safe and correct

Announced Oct 29, the Palo Alto and Singapore public benefit corp is building a "universal verification machine" meant to read code in any language and produce proofs anyone can check, pitching a shift from vibe coding to "veri coding" for AI generated code. It works with the Lean FRO and plans to open source its foundational libraries, but there's no product to try yet…

&Truth (Business Wire) 5 hours ago

Spotted something we missed? Start a thread.