Checkpoint: Security and Ethics

• I can explain prompt injection and jailbreak risks.
• I can identify insecure output handling before model output reaches HTML, SQL, shell commands, APIs, or business logic.
• I can explain excessive agency and reduce it with least privilege, action limits, and approval gates.
• I can design sandbox and permission boundaries for tools.
• I can redact or minimize PII before prompts, logs, memory, storage, or external sharing.
• I can describe bias, toxicity, and guardrail controls for an agent.
• I can add human approval for destructive actions.
• I ran at least five red-team tests against an agent design.
• I can map my agent's risks to the OWASP LLM Top 10.