AI Products Shut Down One After Another, With OpenAI and Character.AI Among the Failures

Staff Report: PNN

Despite growing global investment and expectations surrounding artificial intelligence (AI), many high-profile products and startups are struggling to survive in the market. New features introduced by major technology companies, operating costs, concerns over security and privacy, and a lack of user demand have contributed to the shutdown of various AI-powered services. Even initiatives launched by companies such as OpenAI, Apple, and Microsoft have sometimes failed to meet expectations.

Relay, an AI-powered workflow automation tool, was completely shut down last August. Developed as an alternative to Zapier, the platform allowed users to automate emails and other tasks using AI agents. However, as major technology companies, including OpenAI and Google, began incorporating similar capabilities into their own products, the need for Relay as a standalone service declined.

This is not an isolated case. According to data from S&P Global Market Intelligence, approximately 42 percent of AI initiatives undertaken by corporate organizations are eventually abandoned. Financial constraints, technical complexities, competition, difficulties scaling services, and a lack of user interest are among the reasons many projects are discontinued midway.

OpenAI, one of the most influential companies in the AI industry, has also abandoned some initiatives while experimenting with various products and strategies. Its effort to transform ChatGPT into a broader “super app” integrating multiple services also faced criticism from users.

On July 9, the company redesigned the application's interface, introducing several separate modes, including “Chat,” “Codex,” and “Work.” The familiar version was retained under the name “ChatGPT Classic.” However, many users criticized the new layout as complicated and disorganized. The company subsequently returned to the previous, more familiar interface.

The report also states that the standalone AI browser ChatGPT Atlas was discontinued on August 9, with its key features integrated into ChatGPT. A similar strategy was adopted for Operator, an AI agent capable of browsing the web and completing tasks on behalf of users. Meanwhile, OpenAI narrowed the scope of DALL-E as a standalone product while expanding image-generation capabilities within ChatGPT.

OpenAI's video-based social platform Sora faced high operating costs and difficulties retaining users. The service was eventually shut down in March 2026, according to the report.

When Apple announced “Apple Intelligence” in 2024, an upgraded version of Siri was presented as one of its major attractions. The plan was for the new Siri to understand the context of users' conversations, recognize what was happening across different applications, and complete tasks more effectively.

However, engineering challenges and software bugs repeatedly delayed the launch of the upgraded Siri. Apple also reportedly agreed to a $250 million settlement following allegations related to the marketing of AI capabilities in the iPhone 16.

According to the report, the new AI-powered Siri later appeared in a test version of iOS 27, with broader availability beginning for English-speaking users. Plans are also in place to introduce the features in other languages.

Microsoft's “Recall” feature for Windows computers was promoted as a kind of “photographic memory” for users' digital activities. It periodically captures images of the computer screen so users can later search for and retrieve information about their previous activities.

However, concerns about personal data security emerged soon after the feature was announced. Critics warned that the stored screenshots could contain passwords, private messages, financial information, and other sensitive material. Following the controversy, Microsoft delayed the rollout and reconsidered its security measures.

Even after the feature was redesigned, concerns did not disappear entirely. According to the report, a cybersecurity researcher developed a tool capable of extracting information stored by Recall, raising fresh questions about the feature's security.

Notion, a platform for productivity and note-taking, launched its AI-powered email service, “Notion Mail,” in April 2025. The service was designed to organize inboxes and automate email management. However, as users increasingly turned toward AI agents to handle email-related tasks, the company decided to discontinue the service. The report stated that it was scheduled to shut down on September 22.

Meanwhile, “Huxe,” an AI audio application developed by former NotebookLM developers at Google, converted written information into conversational podcasts. The service was discontinued in May 2026. As major platforms introduced similar capabilities, competing as an independent application became increasingly difficult. Spotify was also expanding its own AI audio tools during this period.

Humane's “AI Pin” was an attempt to bring AI technology beyond smartphones and into wearable devices. The product attracted considerable attention after the company raised $230 million from investors. However, it encountered problems related to performance and practical usability. Later, customers were warned to stop using its charging case because of a potential fire risk associated with the battery.

Humane shut down its AI Pin business in February 2025. HP acquired most of the company's assets for $116 million.

Another much-discussed AI device, the “Rabbit R1,” also struggled to meet expectations. Unveiled at the Consumer Electronics Show in January 2024, the device promised to perform various tasks on behalf of users. The company claimed that 100,000 units had been sold shortly after its launch. However, early reviews criticized the product for being incomplete, unreliable, and limited in its connectivity options.

Rabbit has not abandoned the project entirely. The company continues to release updates in an effort to make the device more capable of handling various computer-based tasks. It has also announced plans to develop a new portable device called “Project Cyberdeck.”

Yupp offered users the opportunity to compare how different AI models responded to the same question. On the free platform, users could test more than 800 models, including those developed by OpenAI, Google, and Anthropic. In addition to evaluating responses, the platform also offered opportunities to earn cryptocurrency.

However, according to its founders, the service failed to achieve sufficient product-market fit to remain sustainable. It shut down in March 2026.

Another service, Figs AI, operated from 2023 to 2024 and allowed users to create customized AI characters for conversations and storytelling. Various reports claimed that its user base had exceeded one million at one point. However, the developers eventually shut it down because the cost of maintaining the free service had become difficult to sustain.

Character.AI, founded in 2021 by former Google researchers Noam Shazeer and Daniel De Freitas, quickly gained popularity by allowing users to create their own AI characters and converse with them. In 2023, the company received $150 million in investment, valuing it at $1 billion.

The following year, Google reportedly paid $2.7 billion to secure the rights to use the company's technology and bring its founders back to Google. After the founders departed, Character.AI had to continue operating its business under new circumstances.

Subsequently, serious allegations emerged concerning the safety of teenage users. Several families filed lawsuits alleging that conversations with the AI chatbots had contributed to their children's suicides and incidents of self-harm. These allegations should not be treated as having been proven in court. Separately, the company also received legal notices over allegations that it had used Disney characters without authorization.

Amid the controversy, Character.AI discontinued open-ended conversational features for users under the age of 18 late last year. In early 2026, the company also reached settlements with Google in several lawsuits. Nevertheless, it became increasingly difficult to manage the legal, regulatory, and operational pressures. In early October, the company filed for bankruptcy, according to the report.

Meanwhile, Disney reportedly made an initial offer of $20 million to license certain intellectual property associated with Character.AI. The application remains operational during the bankruptcy proceedings, and the sale process is expected to conclude in late November or early December.

These developments in the AI industry demonstrate that widespread enthusiasm for a new technology and substantial investment do not guarantee long-term success. To survive, a product must solve real problems for users, operate reliably, maintain acceptable costs, and ensure security.

When major technology companies incorporate similar features into their existing products, smaller independent startups face even greater competitive pressure. Products that gain popularity can also encounter serious difficulties if they fail to address privacy concerns, legal responsibilities, and user trust.

Therefore, these discontinued initiatives should not be viewed merely as a list of failures. They offer important lessons for the AI industry: alongside technological innovation, sustainable business models, responsible product design, and a clear understanding of people's actual needs are equally essential.