How Rubric Keeps You Safe (Iron Dome)
How Rubric Keeps You Safe (Iron Dome)
Every AI-generated response in Rubric passes through a deterministic safety layer — the Iron Dome — before it reaches you.
The problem with AI safety
AI language models are powerful but imperfect. They hallucinate. They make mistakes. A model might estimate a salad at 8,000 calories, or suggest a dish containing peanuts to someone with a peanut allergy.
Most AI-powered apps address this by instructing the AI to be careful. Rubric does both: it instructs the AI to be careful and it verifies the AI’s output with a separate system that does not rely on the AI at all. The AI does not police itself.
What the Iron Dome is
The Iron Dome is a set of deterministic rules that validate every AI response before you see it. It is not AI. It does not “think” or make probabilistic judgments. It matches patterns, enforces hard limits, and checks AI output against your personal safety profile.
The flow is straightforward:
- The AI generates a response.
- The Iron Dome validates it against hard rules and your profile.
- Only then do you see it.
What the Iron Dome checks
| Check | What it does | Example |
|---|---|---|
| Calorie guards | Caps unrealistic nutrition estimates at hard ceilings | A salad estimated at 8,000 kcal is clamped to a reasonable maximum |
| Allergen scanning | Detects allergens in food logging results using your allergy profile | Pad thai flagged for peanut content |
| Hidden source detection | Catches non-obvious allergen sources in common dishes | Pesto flagged for tree nuts; Caesar dressing flagged for fish (anchovy) |
| Coach response scanning | Validates coaching suggestions against your allergy profile | Coach mentions a dish containing your allergen — a safety note is appended |
How warnings appear
The Iron Dome does not silently suppress content. When it detects a potential issue, it tells you.
During food logging: If an allergen is detected in your meal, a colored warning card appears with the allergen name, a description of the risk, and two options: cancel or proceed anyway. You see the warning, you make the decision.
In Coach conversations: If the Coach mentions a food that contains one of your allergens, a safety note is appended to the response identifying the allergen and advising you to verify before acting on the suggestion.
Design decisions that matter
Deterministic, not probabilistic. The Iron Dome uses pattern matching and hard rules. It does not “think” a food might contain peanuts — it matches against known patterns. This makes it predictable and testable.
Post-processing, not prompt engineering. Rubric also instructs the AI to respect your allergies and avoid unrealistic estimates. But it does not rely on that instruction alone. The Iron Dome runs after the AI responds, as a second line of defense.
Non-blocking. The Iron Dome adds warnings — it does not prevent you from seeing AI responses. You are an adult. You get the information and decide what to do with it.
What the Iron Dome does not catch
No safety system catches everything. The Iron Dome covers known allergen patterns and common hidden sources, but novel preparations, regional variations, or unusual ingredient combinations may not be flagged.
The Iron Dome supplements — not replaces — your own knowledge. If you have a severe allergy, always verify food contents independently.
Related Articles
- Allergen Protection — detailed coverage of allergen scanning and hidden source detection
- What the Coach Knows (and Doesn’t) — how your safety profile stays in the Coach’s context
- Not Medical Advice — where Rubric draws the line on medical guidance