Amazon Alexa Plus keeps creepily singing ‘lalala’ for minutes on end

Amazon's premium Alexa Plus service has a glitch that's equal parts creepy and hilarious. Echo speakers running the upgraded AI assistant have started repeating "lalala" for minutes at a time, sometimes chanting it, sometimes singing it, often interrupting actual conversations. When users ask Alexa what just happened, the assistant has no memory of the incident, according to reports compiled by Android Authority.

This isn't a one-off quirk. Reddit threads document the behavior across multiple Echo device types over several weeks, with the most recent reports surfacing just days ago. Users have posted video evidence of their smart speakers stuck in these unsettling vocal loops. Amazon confirmed to Android Authority that it's aware of the issue, though no fix timeline has been shared.

The bug highlights a core tension in the race to ship more capable AI assistants. Alexa Plus launched as Amazon's answer to ChatGPT-style conversational AI, promising more natural interactions and better context awareness. But more sophisticated language models also introduce more unpredictable failure modes. A simpler keyword-based assistant might misunderstand you, but it probably won't start singing nonsense and then gaslight you about it.

What makes this particularly unnerving is the memory gap. The assistant performs an action, users react to it, and Alexa has zero record of what just transpired. That's not just a bug, it's a breakdown in the basic contract of conversational AI: the system should at least know what it just said. It suggests the lalala output might be happening in a different part of the processing pipeline than normal responses.

For context, Alexa Plus is Amazon's premium tier, rolled out to compete with more advanced AI assistants. Users pay extra for supposedly better performance, which makes a bug this visible and weird particularly awkward timing. Google and Apple are pushing their own AI assistant upgrades, and this kind of highly sharable malfunction doesn't help Amazon's positioning.

What this means for you: if you're building products with AI voice interfaces, this is a reminder to instrument logging and state tracking even for edge cases that seem impossible. Reproducible bugs that the system itself can't recall are the hardest to debug. When prototyping voice agents, try this prompt with your AI assistant: "Review the last five outputs you generated and flag any that seem disconnected from the user's input or your stated capabilities." It won't catch everything, but it's a simple quality check for conversational coherence before shipping.