How to Track What Matters to Your AI Models

The AI Intelligence Dashboard can follow the AI models you actually depend on and tell you what changed for them. This guide walks through setting it up and reading what it shows. It takes about five minutes, and it's free.

What you'll need

• A free OnAirToday account. Create one or sign in.
• A short list of the AI models your work relies on: the ones in your product, your team's approved tools, or models you're evaluating.

Step 1: Open your watchlist

Go to the AI Intelligence Dashboard and select Track My AI Models, or jump straight to My AI Watchlist. If you're not signed in, you'll see a short preview and a prompt to sign in. After you sign in, you come back to the same spot.

Step 2: Pin the models you use

Search the model catalog by name, provider or version. Select a model to pin it. You can pin up to 50 models.

You can also pin from any model's page in the AI Model Tracker. Use the Pin to watchlist button at the top of the page.

Tip: pin the exact models you run (for example a specific Claude, GPT or Gemini version), not just the family. Incidents and updates are linked to specific models, so precise pins give you a more precise briefing.

Your watchlist is private. Only you can see or change it.

Step 3: Read your model cards

Each pinned model becomes a card showing:

• The model name and provider
• The latest known version
• The last relevant update
• The number of open security incidents
• The most recent security status
• View Details and a link to the model's risk page

Each card also carries a status label, always shown as text as well as color:

"Status not confirmed" is never counted as open or as resolved. We only show a status the evidence supports.

Step 4: Check what changed since your last visit

When you come back to the dashboard, the What Changed briefing appears above your watchlist. It compares now with your last completed visit and shows four counts:

1. New security incidents
2. Open or investigating incidents (current)
3. Recently resolved incidents
4. New model releases or updates

Below the counts is a feed you can filter by All Changes, Security, Model Updates or Resolved. Each entry shows the model, what changed, the affected version, the old and new status where there is one, the date, a link to the original source, and a link to the OnAirToday details.

A few things to know:

• Your first visit shows a welcome message instead of counts. From then on, each visit has a baseline.
• Your visit is recorded only after the briefing loads. If it fails to load, nothing is skipped next time.
• Browsing the OnAirToday homepage doesn't reset your briefing. Only viewing the dashboard does.
• If nothing changed, you'll see You're All Caught Up! We never pad the briefing with old news.

Step 5: Investigate an incident

Select an incident to open the model's risk page, or browse everything in AI Security & Incidents. For each incident you'll see:

• Category, provider and model(s) involved
• Affected version. If public reporting doesn't confirm it, the page says Version Not Confirmed rather than guessing.
• Status: Reported, Under Investigation, Confirmed, Mitigated, Resolved or Unknown. The status history is kept over time.
• Scope: whether it's a flaw in the model itself, an application or integration issue, a provider security incident, or general research with no confirmed affected model.
• Recommended action and the original source

How to read the scope:

• Model flaw: the issue is in the model's behavior. It may affect anyone using that model or version.
• Application or integration issue: a product built on the model was affected. This does not mean the model family itself is vulnerable.
• Provider security incident: the provider's systems or services were involved, such as an exposed database.
• General research: a technique or finding with no confirmed affected model.

Example: the DeepSeek-V3.2 risk page includes an exposed DeepSeek database that researchers reported. That was a hosted-service failure, not a weakness in the model weights.

We don't mark an incident resolved just because a newer model version exists.

Step 6: Keep your watchlist current

• Unpin a model with the × on its card when you stop using it.
• Add new models as your stack changes, from the dashboard or from any model page.
• Use the AI Advisor or Compare AI Models when you're choosing a replacement, then pin the winner.

Frequently asked questions

Is it free? Yes. You only need a free account.

Can other people see my watchlist? No. Your watchlist and your visit history are private to your account.

Why does an incident say "Version Not Confirmed"? Public reporting often names a model family without saying which version was affected. We show what's confirmed and flag what isn't.

How many models can I track? Up to 50.

Where can I find everything else OnAirToday offers? On the Full Menu.

Start tracking

Open My AI Watchlist and pin your first model. Next time you visit, the dashboard will tell you what changed.