On 13 October, Linus Torvalds replied on the Linux kernel mailing list to a developer who wanted contributors to keep declaring their use of AI tools, calling such rules utterly stupid. The kernel is dropping the tag used for those declarations.
The open-source agent Goose can now be configured with an AI gateway provider declaratively and passes tracing context from its prompts through to the MCP tools it calls, which makes agent runs easier to debug. The release also uses live model metadata for OpenAI reasoning models and caps automatic context compaction at 225,000 tokens by default.
China intends to restrict exports of 23 core AI technologies, including model weights, source code and training methods for systems above 10^26 FLOPs or 2 trillion parameters. Open-source releases would still be cleared quickly once registered, but the move could push Western firms toward American open-weight models.
Google researchers have open-sourced EnvHarness, a framework that adjusts agent-training environments to an agent's weak spots without changing the agent, the environment or the evaluation setup. The researchers say it can make learning both more effective and more efficient.
Many companies want to let staff, including those with little programming experience, build their own applications with AI. Cloudflare has open-sourced a tool called Void to help IT teams keep track of and manage these apps.
AWS has released two open-source projects for building agents: Strands Decider 2B, a small model that makes decisions inside agent workflows, and Strands Box, a sandbox for controlling how agents execute. Together they target agents that call tools, run code and interact with outside systems.