Content Safety

One harmful response on a public agent becomes a headline. Safety layers protect users, especially minors, and shield companies from liability.

Content safety is a bouncer at the door — most guests pass, but harmful content doesn't get served to users.

Scenario

Agent asked for harmful instructions; moderation returns `blocked: violence`; agent responds with policy-safe refusal.

Solution

In Security & Guardrails, apply Content Safety to this scenario: Agent asked for harmful instructions; moderation returns `blocked: violence`; agent responds with policy-safe refusal. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Content Safety (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Content Safety happens in the code.

1safety_settings = [  # key line for Content Safety2    {"category": "HARM_CATEGORY_DANGEROUS", "threshold": "BLOCK_MEDIUM_AND_ABOVE"},3]4response = model.generate_content(prompt, safety_settings=safety_settings)  # key line for Content Safety

• Treating Content Safety as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for content safety

Quick recap — the most important points from this module.