Content safety is now in every Unoblox workspace

Every Unoblox workspace can now turn on Content safety. It checks each request and each reply for sexual, hateful, harassing or dangerous content before it goes through, whichever model you call. The checks cost nothing. Turn it on in Guardrails.

Why it matters

If people type into your product, some of them will try to make it say things you would never publish. Others will paste content you don't want reaching a model, or appearing on your screen. Building filtering into every app takes time, and keeping it consistent across models takes even longer.

Content safety moves that check into the gateway, so every model behind your Unoblox key gets the same protection.

What it checks

Category	What it covers	Text	Images
Sexual content	Pornography, sexually explicit text or images	✓	✓
Hate speech	Slurs and attacks on people for who they are	✓	—
Harassment	Threats, bullying and abuse aimed at a person	✓	—
Dangerous content	Weapons, self-harm and graphic violence	✓	✓

Choose the categories that matter for your product. Both directions are checked: what your users send in, and what the model sends back.

Stop it, or only record it

• Stop it: the request is refused, or the reply is cut off, and the reason is written to the Safety log.
• Only record it: everything goes through, and anything found is written to the Safety log for review.

Recording first is a useful way to see what your traffic actually looks like before you start blocking anything.

Choose how strict to be

Level	Behaviour
Strict	Stops more, and may stop some safe content
Balanced	Recommended for most products
Relaxed	Only stops clear cases

See every decision in the Safety log

Each stopped or flagged request appears newest first, with the reason, the category found, a confidence score, the API key, the model and the request ID. You can search by reason, key, model or request ID, and filter by outcome, category and date (IST).

When a user says "your bot refused my message", you can find the exact request and the reason in seconds.

The fine print

• The checks are free. Content safety itself costs nothing.
• There is a short delay. Checking adds a small amount of time to each request, and streamed replies start slightly later.
• Stopped replies: if a reply is stopped partway, the part already generated is billed like a cancelled request.
• It fails closed: if a safety check is unavailable when a request arrives, the request is refused and you can retry, rather than passing through unchecked.
• Permissions: changing the setting needs permission to manage guardrails.
• It's a filter, not a guarantee. Automated checks reduce risk, and they can miss content or stop safe content. Keep your own product policies and human review where the stakes are high.

Turn it on

1. Open Build > Guardrails in your Unoblox portal.
2. Tick Turn on Content safety for this workspace.
3. Choose your categories, Stop it or Only record it, and a sensitivity level.
4. Save, then watch the Safety log.

Content safety works alongside the data guardrails you may already use: masking for personal, health and card data, IP allowlisting, model restrictions and per-key rate limits.

Open Guardrails or create a Unoblox account.