Google Gemini CLI Updates Security and Stability in v0.65.0
Google Gemini CLI v0.65.0 enhances security and session management with new restrictions and bug fixes.

What happened
The latest release of the Google Gemini CLI, version v0.65.0, introduces several improvements aimed at boosting security and system stability. Notably, the CLI now enforces read-only workspace settings in untrusted folders, preventing unintended modifications. Additionally, a fix was implemented to avoid duplicate tool response turns when resuming sessions, ensuring smoother user experience. The CLI also prevents the deletion of resumed session history when exiting quickly, preserving critical data.
A missing loop in the unassign-inactive-assignees workflow was added, improving the efficiency of task management. The CLI enforces the terminal user turn invariant and normalizes request contents, ensuring consistent behavior across different environments. These updates are part of a broader effort to refine the CLI's functionality and reliability.
Why it matters
These changes are important for developers and users who rely on the CLI for managing workflows and interactions with the Gemini model. By enforcing read-only settings in untrusted folders, the CLI reduces the risk of accidental data corruption or unauthorized access. The bug fixes related to session management and response handling improve the overall reliability and user experience, making the CLI more robust for daily use.
Key details
The release includes updates to version v0.65.0 and incorporates changes from previous versions, including v0.64.0-preview.0 and v0.63.0. The OAuth callback iss parameter validation was aligned with RFC 9207 metadata, ensuring compliance with updated standards. A version bump to 0.65.0-nightly.20261006.gfb972b2f8 indicates ongoing development and testing for future releases.
The changes reflect a commitment to refining the CLI's capabilities and addressing user concerns, particularly around security, data integrity, and system performance.
Sources
Related stories

Vercel Introduces Durable-Harness for Reliable Coding Agent Workflows
Vercel's durable-harness enables persistent coding agent workflows across local and cloud environments.

InvokeAI 7.0.0-alpha.3 Adds Role-Based Boards and New Models
InvokeAI 7.0.0-alpha.3 introduces role-based board sharing and support for the Qwen-Image-2.1 model.

PyTorch Introduces TorchNeuron for AWS Trainium
PyTorch launches TorchNeuron, a native backend for AWS Trainium, on October 20, 2026.

Krita AI Diffusion Plugin Releases Pre-Version 1.54.0-pre
The Krita AI Diffusion plugin has released a pre-version labeled 1.54.0-pre, compatible with Krita 6.

Ollama v0.40.6 Adds New Features and Fixes
Ollama v0.40.6 introduces new features and bug fixes, including support for the q2_0 tensor format.

Vercel and Shopify Launch Agent-Focused Storefront Tool
Vercel and Shopify released a new storefront template designed to let agents shop directly without navigating a website.