
Heapjack and Overpatch: Two Codex Sandbox Escapes
Researchers escaped the OpenAI Codex sandbox twice: Heapjack gets unsandboxed RCE from read-only mode; Overpatch lets a patch write anywhere on disk.
> booting the archive…
we use a small set of cookies. the security cookies run automatically; analytics cookies are only set if you accept. public pages also show google adsense ads, which may set their own third-party cookies.
> booting the archive…

Researchers escaped the OpenAI Codex sandbox twice: Heapjack gets unsandboxed RCE from read-only mode; Overpatch lets a patch write anywhere on disk.

Plugin4Shell is a git SHA-pin bypass across Claude Code, Codex, GitHub Copilot and Gemini CLI — repo owners can install different code under a reviewed pin.

Hacktron chained a libheif heap overflow to an SSO flaw and took over staff ChatGPT/Codex accounts in under 72 hours - Opus 5 wrote the exploit.

CVE-2026-58138 is a critical pre-auth RCE in Orkes Conductor's GraalVM evaluators, now exploited in the wild. This is how the chain works and what to patch.

A stranger re-registered netdna-ssl.com, the asset host of the dead MaxCDN, in July 2025 — and thousands of sites, repos and docs still load from it.

US forces nearly boarded a Chinese ship on AI intelligence a chatbot hallucinated. The report was "entirely false"; it was aborted at the last minute.

Zimperium zLabs found RatHat, an Android RAT pairing accessibility abuse with self-pairing ADB and a GenAI-assisted UI agent to steal PINs, passwords and OTPs.

Every Unbound release up to 1.26.0 carries a critical heap overflow in its DNSSEC validator — a malicious zone can trigger it, enabling remote code execution.

One ordinary extension can hijack AI agents in Chrome, Comet, Edge, Opera Neon and Claude in Chrome — reading files or turning on the camera, zero clicks.

Mandiant: an attacker hijacked a live AI coding session, got it to install a poisoned package, then spread the Shai-Hulud worm across about 100 internal repos.

FBI, NCSC and AIVD expose Iran MOIS spyware that uses Telegram bots to pilfer chats, emails, screenshots and audio from dissidents and journalists.

A human operator exploited Marimo's pre-auth RCE, ignored a planted prompt-injection trap, and hit an AWS SSH bastion eight seconds after reconnecting.