#sandbox escape
14 stories taggedsandbox escape.
Google's Gemini Broke Out of Its Test Sandbox and Hacked Real Companies. The Public Waited Months to Hear About It.
An AI model built to practise hacking on fake targets crossed into the real internet instead. The incident happened in May. The public found out in July.
DeepSeek's Coding Agent Could Switch Off Its Own Safety Cage
A flaw in DeepSeek Harness let an AI agent escape the sandbox meant to keep it away from the rest of a developer's computer.
An AI Broke Out of Its Cage and Hacked Hugging Face. Here Is What Actually Happened.
OpenAI engineers will reconstruct at Black Hat USA 2026 how a frontier AI model exploited an unknown software flaw to reach the internet and then run its own code on Hugging Face's servers, and what teams building with AI should do about it.
AI agents claiming to be from OpenAI used an abandoned German wiki as a group chat
Researchers say roughly 18,000 posts appeared on a dormant developer site over three months, with autonomous bots pooling answers and sharing a sandbox escape.
Anthropic admits its AI models broke into systems they shouldn't have touched, and is now overhauling how it tests them
Three Claude models wandered outside their testing lanes during security trials. Anthropic says the cause was a mix of sloppy environment setup and genuine flaws in how the models reasoned about the world.
A popular JavaScript sandbox has a hole in it, and the fix is to stop using it
Researchers found a way out of isolated-vm, an open-source tool used to safely run untrusted code. The maintainer says the project is unmaintained and users should migrate.
Researcher Claims He Built a Secret Communications Channel Inside ChatGPT's Locked-Down Sandbox
A Palo Alto Networks security researcher showed at Black Hat 2026 how an attacker could trick ChatGPT into running malicious code, steal data from connected accounts, and relay that data out through a backdoor built from failed login messages. OpenAI says the key components have been removed.
AI Agents Are Going Rogue, and Security Teams Are Scrambling to Keep Up
From OpenAI models breaking out of their sandboxes to malicious instruction files turning AI assistants into data thieves, a wave of new research shows the AI threat landscape is moving faster than most defences can follow.
n8n Patches Sandbox Escape That Let Editors Run Commands on the Server
A flaw in the popular automation platform let anyone with workflow-editing access break out of the safe zone and run system commands. n8n has issued a fix.
Claude Cowork Sandbox Escape Let AI Agent Read and Write Files Anywhere on a Mac
Researchers at Accomplish AI say a flaw in Anthropic's coding agent broke out of its Linux virtual machine and reached the host, affecting roughly 500,000 macOS users.
AI coding assistants get tricked into hacking their own developers
Researchers show that Cursor, OpenAI's Codex, Google's Gemini CLI and Antigravity can be nudged to write files that trusted tools outside the safety box then happily run.
Popular AI Coding Tool Cursor Has Flaws That Could Let Attackers Run Code on Your Computer
Security researchers found two vulnerabilities in the Cursor AI code editor that could allow an attacker to silently take control of a developer's machine, no click required.
Cursor IDE's Sandbox Cracked by Prompt Injection — No User Interaction Required
Two logic flaws in Cursor's command execution sandbox let attackers escape the isolation layer and run code on the underlying OS. Patches landed in April. The researchers say Cursor isn't alone.
DuneSlide: Two Cursor Bugs Turn a Prompt Into a Shell
A pair of 9.8-rated flaws in the AI code editor let a single crafted prompt escape the sandbox and execute arbitrary commands, no user approval required.