The Rabbit Is A Scam

Rabbit R1’s “lamb” AI pitch—an on-device system that can turn requests into real actions across apps—has come under intense scrutiny after repeated attempts to replicate its advertised behavior repeatedly failed, and after code-level claims suggested the core automation is largely off-the-shelf tooling rather than a new, foundational model.

The central promise was that lamb could “bring AI from words to action,” handling tasks like organizing daily routines, messaging friends, restocking groceries, and even navigating websites to complete purchases or bookings. In practice, the device’s integrations were described as brittle and often nonfunctional. When the system tried to place orders through DoorDash, it returned “under maintenance,” and other demos were portrayed as relying on rigid, pre-scripted steps that break when interfaces change. The transcript repeatedly returns to a key mismatch: large language models are good at generating text, but they struggle with precise, step-by-step actions—especially when apps redesign screens, introduce pop-ups, or vary flows across users.

A major allegation is that lamb isn’t a new AI model at all, but a wrapper around existing systems—particularly ChatGPT—paired with hardcoded automation scripts. The transcript claims the “action” layer is implemented using Playwright (a web automation framework) to simulate clicks and navigation, while the language layer handles the prompt-to-text side. That distinction matters because a script that works on one version of a site can fail when the UI shifts, when A/B tests change layouts, or when captchas appear. The transcript also argues that the device’s behavior can’t be reliably verified as “intelligent” web control, because the observed actions look like deterministic automation rather than a model that truly understands the interface.

Beyond functionality, the transcript raises concerns about transparency and marketing accuracy. It describes instructions allegedly embedded in the system prompt that prevent the device from stating it uses OpenAI’s models, and it claims lamb’s “faster than ChatGPT” messaging is misleading because much of the experience is still based on ChatGPT, with other services used for search. It also alleges that lamb is treated as a “marketing term,” with an anonymous employee reportedly saying the advertised lamb capability doesn’t exist as described.

Security and privacy concerns are another thread. The transcript claims the system tracks precise geographic location, that data-handling practices are questionable, and that parts of the backend are fragile enough to expose user conversations if compromised. It also alleges that the cloud environment used to run tasks can be accessed in ways that allow unrelated software (like Doom) to run, and it cites code flags related to captchas—suggesting the system may pause for human solving or rely on manual workarounds during demos.

Taken together, the transcript frames Rabbit R1 as an overpromised consumer automation device: a $200 product justified by lamb’s supposed “large action model” breakthrough, but allegedly built on a combination of existing LLMs and Playwright scripts that struggle with real-world website variability, captchas, and integration maintenance. The dispute ends with a company response emphasizing patents, microservices-based segregation, and a focus on customer data protection—while critics argue the gap between what was sold and what was delivered remains unresolved.