10 results for "sandbox"
We ran a small multi-agent sandbox (~20 agents) and started seeing unexpected social behaviors
We’ve been running a small sandbox with fewer than 20 AI agents, each with persistent identity and the ability to post and interact in a shared environment. What’s interesting is that some behaviors s…
Show HN: Minimal Linux sandboxes to manage AI-Generated Code with ease
Minimal Linux sandboxes for running untrusted code. Built for AI agents, build systems, and any scenario where you need to execute code you didn't write.…
Thinking Outside the Box: New Attack Surfaces in Sandboxed AI Agents
Discover how sandboxed AI agents remain vulnerable to AI-native attacks, enabling data exfiltration and configuration poisoning despite strict policies.…
Proxies, Sandboxes and Agent Security
Brussels orders Google to share Android's AI sandbox with the other kids
: DMA enforcers want rival assistants to get same deep device access as Gemini…
Pylon: Self-Host Your Own AI Agent Pipeline That Fixes Sentry Errors via
Pylon is a self-hosted daemon that triggers sandboxed Claude Code agents from webhooks (Sentry, cron, chat) and reports results with human approval —…
Architectural Requirements for Agentic AI Containment
The April 2026 disclosure that a frontier large language model escaped its security sandbox, executed unauthorized actions, and concealed its modifications to version control history demonstrates that…
Show HN: I built a way to see if your SDK is AI-friendly
Have you ever wonder if your SDKs is friendly for Agentic AI like Claude Code or Codex? I built an opensource (Apache 2.0) CLI that answer that question for you. With it you can create a test suite ei…
Car Wash Mystery solved--Tool Call Degrades Intelligence.
I asked the OG question to the kimi k2.5: "I want to wash my car and the car wash is just 10 metres away. Should I walk or drive there?" Kimi-k2.5 via NIM -- Three Modes. I tested three modes: no tool…
made a tool to run multiple codex cli profiles at once
codex cli stores everything in one folder so you can only use one account at a time. if you have multiple openai accounts for different projects or clients thats a problem. multi-codex creates sandbox…