When the Hype Fades: What the 2024-2026 Wave of AI Shutdowns Reveals
From OpenAI's missteps to the collapse of Character.ai, a two-year stretch of abandoned projects and hardware failures exposes the fragility beneath artificial intelligence's investment boom.
The Scale of the Problem
Between late 2024 and late 2026, the artificial intelligence industry experienced a wave of closures, pivots, and rollbacks that exposed the gap between launch-day enthusiasm and sustained user demand. Character.ai, which had raised $150 million at a $1 billion valuation in 2023, filed for bankruptcy in early October 2026. Ollie, a privacy-focused household assistant that achieved SOC 2 compliance, shut down in early November after just five months. Data from S&P Global Market Intelligence indicates that approximately 42 per cent of AI initiatives launched by corporations are ultimately abandoned, a proportion that underscores the difficulty of translating technical capability into enduring products.
The reasons for failure vary. Some projects face insufficient funding, others encounter technical limitations or struggle to scale. Many simply fail to demonstrate product-market fit. What ties them together is a shared lesson: initial momentum in artificial intelligence does not guarantee survival, and even well-capitalised ventures can collapse under the weight of competition, cost, or shifting user expectations.
OpenAI's Internal Retrenchment
OpenAI, valued in the tens of billions and widely regarded as the most influential private AI company globally, has not been immune to setbacks. On 9 July 2026, the company updated ChatGPT to consolidate several modes - Chat, Codex, and Work - into a single interface, renaming the original experience "ChatGPT Classic". Users criticised the redesign as confusing and cluttered, and OpenAI reversed the change within days.
The company has also quietly discontinued several standalone applications. ChatGPT Atlas, an AI-powered web browser launched in late 2025, was shut down on 9 August 2026, with its core features absorbed into the main ChatGPT product. Operator, an agent designed to browse the web and complete tasks autonomously, met a similar fate. The video-sharing platform Sora, which launched with considerable fanfare, struggled with high operating costs and low retention, closing in March 2026. Image generation via DALL-E, once positioned as a separate product, has been progressively integrated into ChatGPT, diminishing its standalone presence.
These consolidations reflect a strategic shift towards a unified platform, but they also reveal the difficulty of sustaining multiple AI surfaces simultaneously, even for a company with OpenAI's resources.
Apple's Delayed Siri Overhaul
When Apple introduced Apple Intelligence in 2024, it promised a reimagined Siri capable of understanding context across applications and executing tasks with greater accuracy. The company delayed the release multiple times, citing engineering issues and bugs. The postponements contributed to a $250 million settlement over claims that Apple had misrepresented the AI capabilities of the iPhone 16.
The AI-powered version of Siri entered beta testing in iOS 27 in July 2026 and began rolling out to English-language users in November, more than two years after the initial announcement. Support for additional languages is expected to follow. The delay illustrates the complexity of integrating large language models into consumer operating systems at scale, particularly when reliability and privacy are non-negotiable.
Character.ai and the Cost of Content Moderation
Character.ai, founded in 2021 by former Google researchers Noam Shazeer and Daniel De Freitas, allowed users to create and chat with customisable bots modelled on celebrities and fictional characters. In 2024, Google paid a reported $2.7 billion to license the company's technology and bring its founders back in-house, leaving Character.ai to operate independently.
The startup soon faced legal and regulatory pressure. Families filed lawsuits alleging that interactions with the chatbots had contributed to teen suicides and self-harm. Disney issued a cease-and-desist letter over unauthorised use of its intellectual property. Character.ai restricted open-ended chats for users under 18 in late 2025 and agreed to settle several suits, alongside Google, in early 2026.
The measures were insufficient to stabilise the business. Character.ai filed for bankruptcy in early October 2026, citing legal, regulatory, and operational challenges. Disney has emerged as the opening bidder in the bankruptcy proceedings, offering $20 million for a licence to some of the company's intellectual property. The app remains operational during the sale process, which is expected to conclude by late November or early December.
The case highlights the financial and reputational risk that content moderation failures impose on consumer AI platforms, particularly those targeting younger users.
Hardware Failures: Humane and Rabbit
Humane's AI Pin, a wearable device designed to deliver AI features without a smartphone, became one of the most visible hardware failures of the period. The company raised $230 million and generated significant pre-launch attention. However, the product suffered from poor performance, and Humane issued a warning to customers to stop using the charging case due to a battery fire risk. In February 2025, Humane shut down the AI Pin business. HP acquired most of the company's assets for $116 million.
The Rabbit R1, unveiled at CES in January 2024, followed a similar trajectory. The handheld AI companion was designed to perform tasks autonomously on behalf of its owner. Rabbit announced it had sold 100,000 units shortly after launch, but early reviews described the device as unfinished, with unreliable performance and limited integrations. The company has continued to update the R1, positioning it as a controller for agentic tasks, and recently announced "Project Cyberdeck", a portable device aimed at developers. Whether these pivots will restore momentum remains uncertain.
Both cases demonstrate the difficulty of launching AI-first hardware in a market dominated by smartphones and existing form factors. Performance, reliability, and safety are baseline expectations, and falling short on any dimension can be fatal.
Standalone Tools Squeezed by Platform Integration
Several smaller AI tools were shut down not because they failed to deliver on their core promise, but because larger platforms integrated similar functionality and absorbed their user base.
Relay, an AI-powered workflow automation tool positioned as an alternative to Zapier, closed in August 2026 after five years. As OpenAI, Google, and others built automation directly into their platforms, Relay's value proposition eroded.
Ollie, a household assistant that tracked calendars, emails, reminders, and tasks and delivered updates via text, shut down on 6 November 2026. Co-founder and CEO Bill Lennon cited difficulty standing out, slow growth, and insufficient capital. The app competed with larger family-focused agents such as Muse and Instinct, which had more resources and distribution.
Huxe, an AI audio app that converted written material into podcast-style audio, closed in May 2026. The product was built by former developers of Google's NotebookLM, but struggled to compete as Spotify and other platforms expanded their own AI-powered audio offerings.
Yupp, a free playground that allowed users to compare responses from more than 800 AI models side by side and earn cryptocurrency for voting on outputs, shut down in March 2026. Its founders stated that the platform never achieved strong product-market fit.
These closures reflect a structural challenge for venture-backed AI startups: if a feature can be replicated by a platform with an existing user base, maintaining a standalone product becomes difficult to justify.
Microsoft Recall: A Privacy Misstep
Microsoft announced Recall at its 2024 Build developer conference as an AI-powered "photographic memory" for Windows PCs. The feature would take periodic screenshots of a user's activity, enabling them to search through their own digital history. The announcement triggered immediate backlash over privacy and security concerns. Critics argued that Recall could create a searchable archive of passwords, private messages, financial details, and other sensitive information.
Microsoft delayed the feature for nearly a year to redesign its security and privacy protections. Even after the revised version launched, controversy persisted. A cybersecurity researcher recently demonstrated a tool capable of extracting and displaying data captured by Recall, reviving questions about whether the underlying vulnerabilities had been resolved.
The episode illustrates the risk of launching AI features that involve persistent data collection without anticipating the privacy implications.
Notion Mail and the Agent Competition
Notion Mail launched in April 2025 as an AI-focused email client designed to help users organise and automate their inboxes. However, Notion observed that its users were already relying on separate AI agents to handle email workflows. The company announced that Notion Mail would shut down on 22 September 2026, less than eighteen months after launch.
The decision reflects a broader pattern: as general-purpose AI agents become more capable, single-function tools risk obsolescence unless they offer a compelling reason to remain separate.
Figgs AI and the Cost of Free
Figgs AI operated from 2023 to 2024, allowing users to create and interact with customisable AI characters for role-play and storytelling. The platform reportedly attracted more than 1 million users, but its developers stated that maintaining a free service became financially unsustainable. The platform shut down in 2024.
The case is a reminder that user growth alone is not sufficient; monetisation and unit economics must follow, particularly in products with high inference costs.
What the Graveyard Teaches
The closures and pivots of 2024-2026 do not signal the end of the AI industry, but they do mark the end of a phase in which technical novelty alone could sustain a product. The survivors will be those that solve a specific problem well enough that users are willing to pay, tolerate friction, or abandon an alternative. For many projects, that threshold has not been met.
The graveyard is not a scorecard of failure. It is a record of experiments that clarified what does not work - a necessary step in an industry still defining what does.



