There are different definitions of Agent Sandboxes circulating, but the strictest defines it as an isolated runtime in which an agent's tool calls execute under constraints that the agent cannot ...

VMs, containers, V8 isolates, WASM and agent sandboxes all promise containment. Here's what each one really stops, what it costs, and where it leaks.

Nine days after GPT-6 Sol, Codex with GPT-6.1 Sol scores 77.7% FuncPass and 34.1% SecPass — within one task of GPT-6 Astra on security, a third faster, and with zero confirmed cheating.

Codex with GPT-6 Sol scores 72.1% FuncPass and 25.1% SecPass for $104 on Azure — 78% cheaper than Astra ($468) — with zero confirmed cheating.

I found a flaw in brig where an agent inside the sandbox plants a symlink resulting in an arbitrary host directory with read-write abilities in and out of the sandbox. Tracked as GHSA-wp6x-29qx-fpr7 ...

A capable frontier model isn’t a controlled system. Here are seven questions security teams should answer before they trust AI coding agents with consequential work.

Trace a change backward from production. The commit, the review, the tests, the dependency it pulled in, the prompt that started it. A person used to sit at every step in that chain. Now they only sit ...

A compromised release of @7nohe/openapi-react-query-codegen runs a dropper on install through three separate triggers, then harvests cloud credentials and republishes itself across npm, RubyGems, and ...

Endor Labs AI SAST now fully supports C. Scans run in the IDE and on the PR, catching vulnerabilities before code ships, for teams writing C faster than release-gated review can keep up. AI assistants ...

I uncovered 14 critical and high severity vulnerabilities, including multiple unauthenticated prompt-injection to RCE chains, across seven AI orchestration platforms. This research was presented at ...