Not long after buying I set Spaghetti detection AI to medium, since it pause my prints so often. However now again I am having that same problem. Maybe its because of the latest update?
Or probably it has something to do about my infill settings? It happens very often on black PLA with 15% gyroid infill. Maybe that looks too much like spaghetti?
It ruined 18h of print already by pausing the print and failing to hold the temp (visible layer shift at paused layer height) 
I would not be surprised if this is partly related to how Bambu has adjusted the spaghetti detection over time.
When the feature was newer, my experience was that Low/Medium gave very few false positives, but also missed some real spaghetti failures. A lot of users then complained that the detection was not aggressive enough unless you used High. My personal impression is that Bambu may have increased the sensitivity, so today Low/Medium can feel closer to what High used to be, while High can become almost unusable on some prints because of false alarms.
Gyroid infill can absolutely make this worse. A 15% gyroid pattern, especially in black PLA, can look messy to the camera because of the shadows, curved lines, reflections, and changing contrast. Depending on model position, lighting, camera angle, infill visibility, and filament color, the AI may interpret perfectly normal infill as spaghetti.
So yes, your infill settings could be part of it, and the latest firmware or AI model update could also have changed the behavior. I do not think it is necessarily a real print failure in your case.
For long prints, I would probably test one of these:
- lower spaghetti detection to Low
- disable it for prints where gyroid infill is very visible
- try another infill pattern if the model allows it
- rotate the model slightly on the plate
- improve/standardize chamber lighting
- make sure the camera lens is clean
The annoying part is that a false pause can still ruin the print. Even if the print itself was fine, the pause can leave a visible layer line, and if the printer does not hold temperature correctly during the pause, the resumed layer can look shifted or poorly bonded.
So in short: I think your suspicion is reasonable. Black PLA + visible gyroid infill + current spaghetti detection sensitivity is a pretty believable cause of false positives.