Litter-Robot 5 Pro's AI mistook two cats for one

Jennifer Pattison Tuohy of The Verge used Whisker's flagship model, the Litter-Robot 5 Pro, at home for six months. Her conclusion was clear: the machine cleans up waste brilliantly, but the AI — the very thing justifying its premium price — simply doesn't work.

What exactly is a robotic litter box

A robotic litter box does the job of scooping cat waste for you. The mechanism is simple: a cat enters a large spherical unit filled with litter, sensors detect its weight and movement, and after the cat leaves, a drum rotates to sift out only the solid waste, dropping it into a sealed drawer below. What used to require scooping twice a day is reduced to emptying a drawer once a week. Most of the odor stays contained inside.

On these fundamentals alone, the 5 Pro excels. It weighs the cat using a built-in scale, can be operated remotely via app, and its LCD screen displays text explaining what action it's currently performing — a major improvement over the cryptic LED indicators on the Litter-Robot 4. It's also nearly silent; the reviewer tested the entry-level Evo model in her daughter's room and was never woken by noise or smell. However, the 5 Pro is bulky, taking up roughly as much space as a side table, and requires a nearby outlet.

Why did Cat ID merge two cats into one

The reason the 5 Pro costs $100 more than the Litter-Robot 5 comes down to two cameras. A front-facing camera recognizes the face of each cat entering the litter box and automatically attaches that visit record to the corresponding cat's profile. Whisker calls this feature Cat ID and states it can distinguish up to five cats.

In practice, after about a week of use, the app prompts a PawPrint setup screen, asking the user to manually select the clearest photos the camera has captured of each cat. In the reviewer's case, every candidate photo was blurry and awkwardly framed. From then on, the machine repeatedly confused the large gray tabby with the smaller, sleeker tuxedo cat, eventually merging the two into what it treated as a single cat.

Whisker explained that poor lighting or weak initial registration images can throw off the camera, and that when the system isn't confident, it falls back on other signals like the built-in scale. In a statement to The Verge, the company said it has "improved the registration process to more selectively choose images used for facial recognition." The company also sells a separate $29 RFID tag for similar-looking cats. But many households don't put collars on their cats, and it seems contradictory for a nearly $900 device to require an add-on accessory just to tell apart two cats that look visibly different to the naked eye.

WasteID and the cameras — advertising versus reality

Another headline feature is WasteID, which is supposed to distinguish urine from feces and prioritize faster cleanup of the smellier waste. Over six months, this feature repeatedly failed to correctly identify waste type, and often couldn't even produce a weight reading for each deposit.

Whisker said the gas sensor used for feces detection on this particular unit was miscalibrated and remotely recalibrated it. When that didn't fix the issue, the company then suspected the scale was off instead. A faulty scale would also explain why both WasteID and PawPrint malfunctioned simultaneously — but fixing it requires replacing the entire unit.

The cameras also fell short of expectations. Marketing suggests both cameras record video, but in reality only the front-facing camera — which captures cats entering — actually records footage. The interior camera only offers a live view, and over six months, the reviewer only managed to capture an actual waste event on camera once. The company says interior recording is coming soon. That said, the cameras weren't entirely useless — when a "cycle interrupted" or obstruction alert came in, the reviewer could check remotely to make sure a cat wasn't trapped inside.

Placement matters more than expected. It took the reviewer about a week to realize the front camera was also capturing people using a bathroom across the hallway. There's an option to record only when a cat is detected, but cats tend to follow their owners into the bathroom too.

What does $8 a month actually buy

The AI health features and video storage are bundled into an $8/month subscription called Whisker Plus. Catching early signs of health issues in aging cats was supposed to be the biggest reason to buy this device — but over six months, the subscription delivered no actionable insights.

A recently added feature called Patterns is at least a step in the right direction, summarizing each cat's day in sentence form — for example, that yesterday Smokey weighed 12.52 pounds, used the litter box once for a total of 0.9 minutes, and that everything fell within normal range. Weight readings are still consistently lower than actual, and waste-type data is still missing. The bigger issue is that Patterns and other core improvements only arrived months after launch — meaning early 5 Pro buyers paid $8 a month for features that weren't working.

Which of the four models should you buy

Model	Price	Camera/AI	Notable limitations
Litter-Robot Evo	$599	None	No litter-level monitoring or night light, small bin, only for small/few cats, no litter hopper support, drawer doesn't close well
Litter-Robot 4	$699	None	LED indicators aren't intuitive
Litter-Robot 5	$100 less than 5 Pro	No camera, includes WasteID	No facial recognition
Litter-Robot 5 Pro	$899	Dual cameras + Cat ID + WasteID	AI features failed in real-world use, requires separate $8/month subscription

The reviewer's recommendations split three ways. If you have a large cat weighing up to 30 pounds, are managing up to five cats, and want the camera mainly as a safety monitor, the 5 Pro can still earn its keep. Otherwise, saving $100 for the camera-less Litter-Robot 5 makes more sense. If you don't care about WasteID tracking and have four cats or fewer, the $699 Litter-Robot 4 — with no AI at all — is the safest choice. Unless space is extremely tight, the reviewer doesn't recommend the cheapest Evo model, citing flimsy build quality and frequent errors, including repeated false alerts claiming the bonnet had come loose. Whisker responded that this, too, was likely a hardware issue specific to that unit.

The hardware is good, the intelligence isn't there yet

In summary, the Litter-Robot 5 Pro is a better waste-scooping machine than its predecessor. The bin is larger, the dome roomier, and the controls more intuitive. The problem is that the $200 premium and $8 monthly fee — justified by the cameras and algorithms — fall apart in the actual home environment. Recognition models that perform well in a lab behave very differently in a real living room, where lighting is dim, cats move quickly, and people wander into frame. If you're buying this product today, you should buy it based on what actually works right now — not on what's been promised for the future.

Editor's Take

Although this review is about a litter box, it's essentially a diagnosis of the entire smart home industry in 2026. Take a machine that already worked fine on sensors and motors, bolt on two cameras and a classification model, raise the price by $200, and then add a monthly subscription on top — this exact pattern is now repeating across doorbells, air purifiers, robot vacuums, and baby monitors. The hardware is already finished; the "intelligence" layered on top is what generates margin. The problem is that the bill arrives before that intelligence has been proven to work.

Ironically, the generational gap ran in the opposite direction. The Litter-Robot 4, which has no AI whatsoever, won't disappoint anyone six months down the line. The 5 Pro, by contrast, created new labor for its users: correcting misattributed visit logs, resetting profile weights, and fixing video logs. The whole point of buying a machine is to eliminate labor — so the moment an AI feature hands labor back to the user, that feature becomes a net negative. Whenever a recognition model built for an appliance is deployed in a real home, the conclusion tends to be the same: 90% accuracy looks great on a spec sheet, but on a device used ten times a day, that's a machine that gets something wrong once daily. And people don't trust a machine that's wrong once a day.

There are three lessons for domestic teams to take away. First, if individual identification is going to be a flagship feature, having users manually pick registration photos is already a losing design — if every photo the user can choose from is blurry, that's a pipeline problem, not a user problem. Second, if AI features are going to be bundled into a subscription, billing should only begin once those features are confirmed to actually work. Shipping core improvements months after launch while charging fees the whole time erodes brand trust far more than it invites refund requests. Third, the most alarming detail is that a single miscalibrated scale simultaneously broke both facial recognition and waste classification. When multiple AI features all depend on the same single sensor, the moment that sensor is wrong, the entire product starts lying. A design that detects its own sensor faults and flags the affected feature as "undetermined" and halts is always better than one that fills in the gaps with guesses.

It's not hard to predict what happens next. Whisker will likely expand Patterns and roll out interior camera recording in an attempt to quiet early buyers' complaints. But even after those updates ship, the core lesson of this product won't change: you shouldn't buy hardware based on promised future updates. What works the day you open the box is what you actually bought.