Claude's prompt injection attack surface extends far beyond chat. Skills execute with system-prompt authority. Web content flows into agent reasoning without trust boundaries. MCP servers bridge ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results