Recursive Read-Only Mounts With mount_setattr Need a Verified Boundary
Why recursive mount attributes, descriptor identity and alias review all matter when building read-only sandboxes.
Tag archive
Why recursive mount attributes, descriptor identity and alias review all matter when building read-only sandboxes.
The Imperative of Least Privilege: Securing GitHub API Access In the fast-paced world of...
To detect AI agent actions outside task scope, check each tool call against a declared task contract before it runs, and block anything unnamed.
A practical audit of Vault PKI role constraints and issuance boundaries.
How Elasticsearch API key permissions intersect with the creating principal.

MCP prompt injection: cut what an agent can do with narrow tools, a minimal token and human approval, then test the guardrails and trace every call.
CVE-2026-114625 makes every VeloCloud Orchestrator login a potential escalation path, including read-only tenant accounts. Here is how to re-scope account inventory.
How to read a cloud policy element by element, which wildcard combinations produce escalation or public access, and the controls that cap blast radius.
BeyondTrust's Phantom Labs found that OpenAI Codex passed branch names into a shell without sanitizing them, so a single semicolon could exfiltrate the task's GitHub OAuth token. The bug is fixed, but agents still commonly hold over-scoped, long-lived credentials, and that is what made it dangerous.
A repeatable Kubernetes RBAC audit that resolves aggregated roles and ranks grants by privilege-escalation potential.
Short answer: Give the admin console its own named API key with the narrowest scopes it needs, and...
Least privilege for AI agents in practice: how to scope tools, credentials and filesystems, with the Replit and Amazon Q incidents as evidence.