Unverified60% confidenceFactExact time
OpenShell uses formal verification to handle policy changes: when a policy is modified, it first uses mathematical methods to verify which new access permissions the change would grant, and high-risk changes are always blocked pending human review, so the agent cannot escalate its own permissions.
2
Sources
60%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
NVIDIA Open-Sources OpenShell: Putting a Safety Cage Around AI Agents
bilibili岩石与宇宙的家10/25/2026
Related Entities
Related Claims
UnverifiedUsing OpenRouter's preset configuration—setting the only parameter to lock the provider, setting allow_fallbacks to false to disable fallback, and disabling the reasoning feature—can ensure consistency and traceability of test results.70% similarUnverifiedThe new model has stricter safety policies, with penetration-testing-type requests blocked outright, requiring users to go through a verification process to access expanded capabilities.68% similarUnverifiedDifferent verification strategies are applied based on operation risk level: trust the tool response for read-only or reversible operations, use read-back checks for important write operations, use idempotency keys plus retries plus read-back for financial or irreversible operations, and supplement batch or background tasks with external monitoring and alerting.67% similarUnverifiedAdopting a pilot rather than a full-replacement strategy can verify the feasibility of open-source solutions while controlling the risk of failure, and can strengthen the government's bargaining leverage in negotiations with Microsoft.67% similarUnverifiedUnder the Confirm before run policy, read-only commands execute directly, change-type commands require confirmation, and high-risk commands require secondary confirmation.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1103868API
curl https://kongchang.com/api/v1/knowledge/claims/1103868MCP
get_claim(id=1103868)