AI Rivalry Shifts to Agent Runtimes as Permission Gaps Persist

Competition in artificial intelligence is moving beyond foundation-model performance toward agent runtime frameworks that coordinate models, tools, data and access rights. Mozilla says control over how agents plan and execute tasks could shape enterprise adoption, ecosystem influence and user safety, making runtime governance an increasingly important battleground as AI systems gain the ability to act rather than merely generate responses.

Mozilla’s latest report says interoperability protocols including MCP and A2A are helping models, tools and agents work across systems. Yet the industry still lacks a common, cross-framework permission standard for higher-risk actions such as writing data, changing settings or executing transactions. The report also warns that repeated approval prompts may cause “consent fatigue,” weakening oversight. It specified neither a launch date for a unified standard nor any associated monetary figure.