Anthropic researchers found AI agents can clash, collude, and coordinate in unexpected ways, raising new questions about whether today’s safety tests capture the risks of multi-agent systems.
Published: 2026-08-13T18:28:14+00:00
Source: techcrunch.com