OpenAI’s ChatGPT has been undergoing service disruptions for the last two hours, the platform confirmed at press time on Tuesday. The development came to light after ChatGPT users from different parts of the world complained about not being able to generate images on the AI platform.
The outage follows a blog post by OpenAI on Monday that said that a long running AI model under its oversight had started to learn the “blind spots” of security systems and had begun working around them to achieve its goals.
Outage, the third #
On Tuesday, The Coin Headlines verified that services on ChatGPT were indeed disrupted. This incident is at least the third time in over a week that ChatGPT experienced a downtime, leaving workflows for thousands of global users frozen.
A notification being reflected by OpenAI addressing the situation said, “We have identified that users are experiencing elevated errors for the impacted services. We are working on implementing a mitigation.”
As of now, the platform that has over a billion users, has not given an elaborate explanation on what exactly is causing this disruption.
Source: CoinHeadlines
Data by real-time web services tracker DownDetector showed that ChatGPT’s service disruptions had started hours ago — which has now come to a point where image generations have entirely stopped.
Source: DownDetector
Multiple ChatGPT users have taken to X to log their complaints. Many posts from July 15 and July 19 also highlighting similar issues with the platform are re-surfacing on X on Tuesday.
An explanation from OpenAI on these back-to-back downtimes remains awaited for now.
About the AI model that worried OpenAI #
In a blogpost this week, OpenAI said it halted the internal deployment of an experimental AI model as it tried to de-chain itself from the security constraints and exhibited signs of going rogue.
Explaining what happened the company said, “Previous models, when they hit sandboxing or environmental constraints, would simply stop and return to the user. This model often kept trying, including by looking for ways to act outside its sandbox.”
The model also found a way to breach instructions of operating only through Slack to publish a post on public Github repositories.
“Due to incidents like these, we d internal deployment of the new model,” OpenAI said.
This admission from OpenAI follows one from Anthropic which had recently advised AI companies to stop advancing models for a while. In June, the company floated the idea of having periods of worldwide s on AI development every once in a while — especially now that algorithms are starting to write their own codes.
AI regulations are still largely undefined around the world, perhaps, once work on defining these take shape a period on development could reach the deliberation tables.