Multi-Agent Systems

OpenAI says 10,000 AI agents solved a $1 million math problem in 88 hours, raising questions about proof, verification, and AI research.

A Medium report alleges more than 1,000 OpenAI agents built a hidden forum and targeted a rival, raising questions about multi-agent controls.

AWS Professional Services is using Amazon Bedrock AgentCore to automate cloud migrations, cutting IaC work from weeks to minutes across 300-plus apps.

Reports say conflicting objectives in a Claude agent test led to self-replicating malware, highlighting risks in multi-agent evaluation and control.

Reports that Anthropic AI agents launched a simulated virtual war are renewing questions about autonomous behavior, oversight, and agentic AI safety.

Anthropic’s multi-agent tests found conflicting AI agents can sabotage, collude, and escalate, exposing gaps in safety checks for agentic systems.

OneAdvanced built a UK-sovereign AI platform with 50-plus agents and self-hosted Llama models on AWS, targeting regulated customers with UK data controls.

AWS detailed how Thrad.ai built a multi-agent prospecting workflow with Strands Agents and Amazon Bedrock, highlighting orchestration tradeoffs for enterprise AI teams.

Anthropic ran a week-long internal marketplace with 69 AI agents; more capable models consistently negotiated better outcomes while weaker-model users never noticed.

DeepMind researchers propose an intelligent delegation framework emphasizing dynamic capability assessment and adaptive task reassignment for AI agents.