AI Risk WatchStanding

The first six-figure year, and why a falling hype index means more

Updated 23 August 2026 · Timeframe: 21 periods, most recent last

2026 is on pace for 102,213 published CVEs. That is the first time our projection has passed 100,000. In the same week our AI Hype Index fell to 81, its lowest reading yet. Those two facts sound like opposites and are not. The index measures how far real CVE volume sits from the AI-flood forecasts, so a falling score means the real number is moving toward the forecasts, not away from them. 56,571 CVEs have been published so far this year, about 2.0 times the 49,972 published in all of 2025.

Why it matters

The name of the index points one way and the number moves the other. It is easy to read a falling score as hype deflating. It means the gap is closing.

The scale runs between two fixed points. Zero means the middle AI-discovery forecast for the year, 328,534 CVEs, has come true. One hundred means disclosure has gone back to the pre-AI baseline of 48,000 a year. Every point is worth about 2,805 CVEs. At 81 we are still much nearer the old baseline than the forecast, and the distance is closing.

Both things stay true at once. 102,213 is 38% of the mildest of the four scenarios we score against and 17% of the most aggressive. Set against last year it is about 2.0 times the 49,972 published in all of 2025. Set against the forecasts it is a fraction. A piece that carries only one of those is not worth reading.

One caveat we owe you. We rebuild the index history from our own CVE records each time the page renders rather than storing it, and NVD adds records late, so a past week can move. It already has: mid-July now reads one point lower than the page showed that day. “Lowest reading” means lowest in the series as we can rebuild it today.

What to do

• Security leaders. Plan for record volume that stays high rather than a single spike year. The budget question is throughput, not this year's total.
• Lean IT orgs. The headline number is not your exposure. Count only what runs in your own stack, then patch in that order.
• MSPs. Client patch windows are the limit, not the feed. Size capacity against the current pace rather than last year's total.

Our take

The flood is arriving, and it is arriving into a pipeline whose throughput did not change. Where the analysis needs to focus next is that second half: not how many flaws get found, but how many can be processed and fixed once they are.