Runtime Passports and MicroVM Sandboxes Harden the Agent Stack
ServiceNow's new AI Passport gives agents cryptographically signed credentials checked at runtime, with a machine-speed kill switch enforced through NVIDIA's Open Agent Safety Platform, while Perplexity open-sourced SPACE, a Firecracker-microVM sandbox with zero escapes in 108 red-team runs. NVIDIA also expanded
Highlights
- ServiceNow shipped AI Passport, a cryptographically signed runtime credential that verifies a first- or third-party agent's scope at execution time and can throttle, redirect, or kill it at machine speed when it drifts — with NVIDIA as design partner enforcing it through an Open Agent Safety Platform (ServiceNow).
- Perplexity open-sourced SPACE (Apache 2.0), a Rust Firecracker-microVM sandbox for agent code execution: ~60ms median spin-up, 3–5x faster than the commercial provider it replaced, and 0 escapes in 108 red-team attempts (Crypto Briefing).
- NVIDIA expanded Nemotron 3.5 with an open 120B MoE Super VL model built for always-on enterprise agents, plus two Bolt models tuned for local and edge hardware (The Decoder).
- Governance moved from audit trails to prevention: the new ServiceNow stack rebrands Control Tower as a Super Intelligence Control Tower with shadow-AI detection, 400+ connectors, and cost management folded into ITAM (SiliconANGLE).
Key Signals
-
Agent identity becomes runtime enforcement, not just a badgeOct 15, 2026
After this week's wave of agent-identity launches, ServiceNow's AI Passport pushes past issuance into graduated runtime control: agents carry signed credentials grounded in CMDB/GRC context, and the Control Tower can revoke session tokens, restrict tools/endpoints, or revoke credentials outright (Help Net Security). NVIDIA is the launch partner, enforcing passports via its OpenShell secure runtime and Sentry framework (Constellation Research). AI Passport, shadow-AI detection, and AI Cost Management are slated for Q4 2026.
-
Sandbox isolation gets an open, battle-tested reference implementationOct 15, 2026
Perplexity's SPACE isolates each agent run in a Firecracker microVM rather than a shared-OS software boundary — the kind of deep isolation operators have wanted since MCP tool-call vulnerabilities became front-page. It's been in production since June 2026 powering Perplexity Computer, and the red-team numbers (0/108 VM escapes; a network-gateway bypass found and patched) are published rather than asserted (Crypto Briefing).
-
Open models aimed squarely at nonstop agent workloadsOct 15, 2026
Nemotron 3.5 Super VL activates ~12B of 120B params per token and ships with weights, data, and recipes, targeting instruction following, tool use, and multistep workflows at roughly a fifth of a flagship's parameter count. Bolt 74B (24GB) and E4B (4GB, due Oct 23) take the same stack down to local and edge agents (The Decoder).
Why It Matters / What To Watch
-
The agent control plane is consolidating around signed identity + hard isolation.
- If you run agents across clouds, evaluate whether a runtime passport model (scope verification + machine-speed kill switch) fits your existing IAM/GRC before vendors lock you in (Help Net Security).
- Treat SPACE as a reference for your own sandbox threat model — microVM isolation plus a hardened egress gateway, not software boundaries alone (Crypto Briefing).
-
Open weights are now a real option for always-on and on-prem agents.
- Benchmark Nemotron 3.5 Bolt against hosted APIs on your tool-use and workflow traces; the economics of local inference shift once a 4–24GB model handles the routine hops (The Decoder).
- Watch serving stacks catch up: AWS Bedrock AgentCore continues to iterate its runtime, Gateway, and MCP-server plumbing, so portability of agents across these control planes is the open question (AWS).
Quick Links
- ServiceNow advances its operational harness for enterprise AI with AI Passport - ServiceNow Newsroom
newsroom.servicenow.com - ServiceNow gives AI agents a passport, and limits on what they can do - Help Net Security
helpnetsecurity.com - ServiceNow launches AI Passport to ride along with NVIDIA's Open Agent Safety Platform - Constellation Research
constellationr.com - Perplexity open sources SPACE, its sandbox for AI agents - Crypto Briefing
cryptobriefing.com - Nvidia's Nemotron 3.5 Super VL is an open model built to run enterprise AI agents nonstop - The Decoder
the-decoder.com - ServiceNow adds AI agent controls and employee automation tools - SiliconANGLE
siliconangle.com - Release notes for Amazon Bedrock AgentCore - Amazon Web Services
docs.aws.amazon.com