Granite Guardian vs Lakera Guard (Check Point AI Guardrails)

Granite Guardian

One ungated Apache 2.0 model judges harm, jailbreaks, RAG groundedness, function-call errors and custom criteria, with signed weights and published evaluation code. It is trained and tested on English only, each call checks one criterion, the 4.1 prompt format differs from 3.x, and IBM's watsonx.ai lists only the deprecated 3.0 model.

Lakera Guard (Check Point AI Guardrails)

OpenAI message format in, including tools, tool calls and tool results. Prompts and outputs are stored for the dashboard by default.

Granite Guardian

1. Append the <guardian> block as the final user message, with the mode line, ### Criteria: and ### Scoring Schema:. Copy the strings from the model card, because no package builds them
2. Use the no-think instruction for gating and parse <score>. Think mode writes a reasoning trace first, and the card's examples allow up to 2,048 output tokens
3. Treat yes as the criterion being met, which for built-in criteria means the risk is present. Treat a missing <score> tag as a failed check
4. Pass retrieved text through documents= and tool schemas through available_tools= in apply_chat_template, not inside the message text
5. Under Ollama, set num_ctx in the request options. IBM's docs say the default context is short and long requests are truncated

Lakera Guard (Check Point AI Guardrails)

1. Start the project in Detect mode and calibrate before Enforce. In Detect mode flagged is always false
2. Send the whole conversation, but expect only the last user, assistant or tool turn to be scored
3. Ask for breakdown=true and treat the confidence levels as a dial, not a boolean
4. Turn off prompt logging in General Settings if the dashboard shouldn't hold user content
5. Pin eu.api.lakera.ai or us.api.lakera.ai rather than api.lakera.ai when residency matters