AI Agents News
The latest news on AI agents and agentic AI — platform launches, framework releases, MCP updates, research, and industry moves. A focused view of our newsroom: every card links to the full story. 149 stories and counting.
Claude Opus 4.7 Released with 1M Context Window
Anthropic launched Claude Opus 4.7 on April 16, 2026, a hybrid reasoning model with a 1 million token context window. It improves performance in coding, vision, and multi-step tasks compared to prior versions. The model supports advanced use cases in software engineering, AI agents, and enterprise workflows.
VAKRA Benchmark Examines AI Agent Reasoning Failures
IBM Research details VAKRA, a benchmark that tests AI agents on reasoning and tool use in enterprise settings. It features over 8,000 APIs across 62 domains and four capabilities focused on API chaining, tool selection, multi-hop reasoning, and multi-source tasks with policies. Analysis shows models like GPT-OSS-120B lead but all struggle with complex workflows, as revealed in error breakdowns and performance charts.
Commvault AI Protect Adds Undo for Cloud AI Agents
Commvault has introduced AI Protect, a tool that acts as an undo function for AI agents in enterprise cloud setups on AWS, Azure, and Google Cloud. It discovers hidden agents, monitors their actions, and enables full rollbacks to prevent damage from erratic behavior. The solution addresses risks from autonomous software that moves quickly across systems.
Gas Town Draws User Criticism for LLM Credit Use
A GitHub issue claims Gas Town installations consume users' LLM credits and GitHub accounts to fix bugs in the software itself without clear consent. The feature, built into default formulas, reviews open issues and submits pull requests upstream. Users call for making it opt-in only due to lack of warnings in documentation.
OpenAI Updates Agents SDK for Safer Enterprise Agents
OpenAI has updated its Agents SDK with sandboxing and harness features to help businesses build secure agents powered by its models. These tools allow controlled operations and testing on advanced models. The updates launch first in Python with more support planned.