AI agent access and security exposure

Anthropic gives Claude Code a live window into a running iOS app

Two additions push a coding agent past editing source, into watching a running application and auditing a repository. Both rely on access that two documented flaws have already abused.