Big Tech AI Lacks Outage Compensation Even When Services Go Down

ChatGPT, Claude, Gemini
43 Outages Over 3 Months
“Consumer Terms Must Specify Duties”

Photo Image
AI-generated image

More than 40 service disruptions lasting 30 minutes or longer were recorded over the past three months across generative artificial intelligence (AI) services provided by OpenAI and Anthropic, according to findings confirmed on Sunday. Although generative AI services are widely used in work and daily life, outage compensation policies for individual users remain deficient under terms of service, prompting calls for improvement.

On Sept. 15, an analysis by Electronic Times of service disruptions posted on each company's official site over the past three months (June-August) showed a total of 43 outages lasting 30 minutes or more across three major Big Tech AI services with high usage rates in South Korea: ChatGPT, Claude, and Gemini.

By service, OpenAI's ChatGPT accounted for 12 cases, Anthropic's Claude recorded 30 cases (including individual model outages), and Google's Gemini had one case.

OpenAI primarily experienced connection failures or functional disruptions in specific features such as chat and image generation. Anthropic suffered from errors in specific models expanding broadly into code and co-work features, as well as degraded chat and conversation performance.

Google Gemini recorded only one outage, but it was the longest single disruption identified during the three-month period. In June, the outage lasted 6 hours and 55 minutes, causing approximately 50% of prompt transmissions on web and mobile platforms to fail.

Even when outages occur, compensation measures for paid subscribers (B2C) remain nonexistent. None of the three services include outage compensation principles in their consumer terms of service.

When a disruption occurs, OpenAI determines necessary measures by comprehensively reviewing the scope, impact, and relevant national laws. Based on these standards, credit compensation was provided once in South Korea during an outage in June last year, but outage compensation is not explicitly stated in the service terms.

Anthropic does not specify outage compensation in its terms for paid Claude plans, nor does it have a record of providing separate compensation.

While setting no compensation measures for general paid users, these companies exhibit discriminatory behavior by entering into Service Level Agreements (SLAs) for enterprise clients. Enterprise clients can receive compensation under negotiated SLAs when service disruptions or other issues occur.

“ChatGPT is a service used by more than 16m people monthly in South Korea, and because it is utilized for work and critical tasks, appropriate compensation should be provided as users suffer losses during outages,” an industry official said. “Duties should be considered under the Basic AI Act regarding safety obligations and the Telecommunications Business Act regarding value-added telecommunications service stability.”

· This article was translated using AI and was published after final review by the reporter.