"How Many Times This Year? Repeated Generative AI Model Errors... How Are Paid Users Responding?"

OpenAI Has Not Disclosed the Cause of the Error

Frequent Outages for Anthropic's Claude Amid Surging Usage

"Compensation Systems for Paid Users Should Be Reviewed"

Situations where users experience inconvenience due to errors occurring in generative artificial intelligence (AI) models are repeatedly taking place.

According to industry sources on May 12, disruptions on OpenAI's ChatGPT have occurred for the third time this year. As reported on OpenAI's official status page on May 11, errors occurred on ChatGPT and Codex around 10 a.m., leading to user inconvenience. Issues arose when uploading files to ChatGPT or working on Codex Cloud. For example, when users tried to upload attachments such as photos or PDF files to the ChatGPT chat window, the system would remain stuck in an infinite loading state. OpenAI applied mitigation measures and monitored the recovery, and was able to normalize the system around 2 p.m., about four hours after the error occurred. OpenAI did not disclose the cause of the error.

At the end of April, the ChatGPT service was suspended, and some users were unable to log in to platforms such as Codex. That incident was caused by an overload in the internal system that checks ChatGPT's login and authentication. OpenAI later stated, "We are strengthening the system to improve the availability of ChatGPT during periods of high system load or when parts of the infrastructure are down."

In early February, access disruptions in the United States led to more than 13,000 reports on Downdetector, a website that tracks service outages. Users experienced delayed responses and login failures, among other inconveniences.

Not only OpenAI but also Anthropic has experienced several service disruptions this year. In particular, Claude had three outages just last month. Users were unable to access Claude and faced inconvenience. Due to the surge in usage and the burden of infrastructure costs, Anthropic revamped its pricing model to a usage-based system. The basic fee for the enterprise plan was lowered to $20, and users are now charged based on actual usage. To respond to the increasing demand, Anthropic has also expanded strategic partnerships with Google and Broadcom, securing 3.5 GW (gigawatts) of AI computing capacity starting next year.

Lee Jaesung, professor of AI at Dongguk University, analyzed, "The file upload malfunction was likely caused by cloud-related network errors." He added, "Accumulated usage records on ChatGPT improve work efficiency, but if users switch to another AI model due to errors, their past records do not remain, which may cause a decrease in productivity. For people using paid services, there could be periods where they were unable to access the service due to errors, so it would be worth checking with organizations such as the Korea Consumer Agency to learn about compensation policies."

This content was produced with the assistance of AI translation services.