The AI graveyard: a running list of projects and startups that didn’t make it
Relay was an AI tool that automated work, similar to Zapier. It shut down completely in August. The product used AI agents to help people automate email and task-related work. But larger companies, including OpenAI and Google, began adding similar features to their own tools. As a result, the five-year-old company Relay struggled to give people a reason to keep using its product on its own.
More recently, chatbot platform Character.ai filed for bankruptcy, just five years after it launched.
Many AI projects also fail inside the companies that create them. According to S&P Global Market Intelligence, about 42% of AI projects are eventually abandoned by the companies behind them. The reasons differ. Projects may not get enough money, face technical problems or strong competition, struggle to grow, or fail to attract enough users.
As a result, the list of failed AI projects keeps growing. Below are some of the best-known AI products, startups, launches, and projects that have shut down, changed direction, or fallen far short of expectations. This is not just a list of failures. It is also a record of what these attempts might teach the rest of the industry.
OpenAI has had several setbacks
Even the biggest companies can have failures. OpenAI, one of the world's most valuable privately owned companies, has had its own share of projects that did not work out. One recent example was its attempt to turn ChatGPT into a broader “super app.” On July 9, OpenAI changed the app to bring several experiences together. These included separate “Chat,” “Codex,” and “Work” modes. The company renamed the familiar version of the app “ChatGPT Classic.”
The new design quickly faced criticism. Users said the new layout was confusing and crowded. OpenAI reversed the change soon afterward and brought back the familiar ChatGPT layout.
This was not OpenAI's only attempt to put many features into one product. The company has also closed several separate apps and moved their features into ChatGPT. That approach has worked better than the redesign. ChatGPT Atlas, a separate AI-powered web browser, lasted less than a year before it was discontinued on August 9. Its most useful features were added to ChatGPT.
The company did the same with Operator, an AI agent that could browse the web and carry out tasks for users. OpenAI has also added more image-making features directly to ChatGPT. This has reduced the need for DALL-E to remain a separate place for creating images.
OpenAI's video-sharing platform, Sora, also struggled. It faced high running costs and difficulty keeping users. The company shut it down in March 2026.
Apple's Siri AI
When Apple introduced Apple Intelligence in 2024, a better version of Siri was one of its main features. Apple promised that Siri would understand context, know what was happening across apps, and complete tasks well.
However, Apple repeatedly delayed the new Siri. The company pointed to several problems, including technical difficulties and software bugs. These delays eventually helped lead to a $250 million settlement over claims about how Apple advertised the AI features of the iPhone 16.
The new AI version of Siri finally appeared in the iOS 27 beta this past July. The updated version began rolling out to English-speaking users first this month. Support for other languages is expected to come later.
Microsoft's Recall
Microsoft announced its Recall feature at its 2024 Build developer conference. It was presented as an AI-powered “photographic memory” for Windows computers. The feature would regularly take screenshots of what users were doing, allowing them to search through their past computer activity later.
As expected, the announcement quickly caused concern about privacy and security. Critics said Recall could create a searchable record of very private information, such as passwords, personal messages, financial details, and other activities. Microsoft delayed the feature for almost a year so it could rethink how to protect users' privacy and security.
Even after Microsoft redesigned Recall, concerns continued. A cybersecurity researcher recently built a tool that could extract and show data captured by Recall. This raised new questions about whether Microsoft had really fixed the main security problem.
Notion Mail
Notion Mail launched in April 2025. It was an AI-based email product meant to help people organize their inboxes and automate email tasks. But the company noticed that users were choosing separate AI agents to manage their email instead. Because of this, Notion Mail is scheduled to shut down on September 22.
Humane AI Pin
The Humane AI Pin is one of the best-known failures in AI hardware. It was designed as a wearable device that provided AI features without needing a regular smartphone. Humane raised $230 million from investors and attracted a lot of attention.
But the product had serious performance problems. Things got worse when Humane warned customers to stop using its charging case because the battery could pose a fire risk. Humane shut down its AI Pin business in February 2025. HP bought most of the company's assets for $116 million.
Rabbit R1
The Rabbit R1 was another AI device that attracted a lot of attention. It was introduced at CES in January 2024 and was designed to act as an AI helper that could do tasks for its owner. Rabbit said it had sold 100,000 units soon after launch.
However, strong early sales did not mean the product worked well. Early reviews described the device as unfinished. It often did not work reliably and could connect to only a limited number of useful services.
The company has not given up. It has continued to update the R1 and now presents it as a controller for computers that can carry out tasks using AI agents. Rabbit also recently announced a new hardware project called “Project Cyberdeck.” The goal is to build a portable device designed for vibe-coding, or creating software with the help of AI.
Huxe
Huxe was an AI audio app created by former developers of Google's NotebookLM. It turned written information into audio conversations that sounded like podcasts.
The company shut down in May 2026. One reason was that larger platforms began offering similar features to many existing users. For example, Spotify has been adding more AI-powered audio tools of its own.
Yupp
Yupp was a free AI testing platform. It let users compare answers from hundreds of AI models side by side. Users could vote on the answers and earn cryptocurrency through the platform.
At its busiest, Yupp let people try more than 800 AI models, including models from OpenAI, Google, and Anthropic. But the founders said the product never found enough demand from the right users to survive. The platform shut down in March 2026.
Figgs AI
Figgs AI operated from 2023 to 2024. It let users create and interact with AI characters that they could customize for role-playing and storytelling. The platform reportedly attracted more than 1 million users. But its developers said the cost of keeping the service free became too high to manage, so they shut it down.