Claude dynamic workflows and the Anthropic Cyber Mission

A Managed Agents agent can now write a workflow (a program that runs many agents in phases and merges their results) and run it server-side in the background. The target is large jobs, such as reviewing hundreds of documents. You turn it on with the agent's multiagent field and track progress through workflow_run.* events.

It can take over fan-out work without you writing orchestration code. That may cut the effort behind bulk document or code review pipelines. The notes did not cover cost or agent-count limits, so try it on a small job first.

Anthropic launched a long-term cyber program for defenders. The main piece is OSS Scanner, which gives open-source projects periodic free security scans from its strongest models. Reports include a proof of concept and a suggested fix, but they are model-generated and not human-reviewed. A Critical Infrastructure Defense Program (CIDP) and an expanded Cyber Verification Program also launched.

Maintainers can enroll, though the program is meant for projects that can handle the volume of findings. The vendor expects a true-positive rate above 90%Vendor claim. Nobody has measured it independently yet.