DeepSeek servers suffer global outage, mysterious Harness accounts emerge simultaneously

So early in the morning, DeepSeek is down??

I rushed to check the official status page, and it turns out that the problem is not unique to me.

DeepSeek has posted a red outage notification:

Partial outages are currently occurring on DeepSeek web and API services. The root cause has been identified and remediation is in progress.

The affected services currently include DeepSeek Chat service, DeepSeek V4 Pro API and DeepSeek V4.1 Flash API.

What on earth happened??

Chat and APIs both stop working, netizens: Is a new model coming?

According to the accident record released by DeepSeek official, this outage first occurred at 8:11 a.m. Beijing Time on October 15.

The official initially stated that the team was investigating the problem and locating the root cause.

Subsequently, the accident status was further updated to "root cause identified" and entered the remediation phase.

As of the latest official progress update, this outage has lasted for about 1 hour.

From user feedback, the situations people encountered are quite different:

Some people never received a reply after sending a message, some were stuck on the loading interface all the time, and others directly encountered prompts such as "Server busy" and "Messages are sent too frequently, please try again later".

The overseas service monitoring account Status Is DOWN also quickly issued a reminder, stating that DeepSeek is experiencing anomalies and asking if other users are also affected.

The third-party outage monitoring platform StatusGator shows that 86 related fault reports have been received in the past 24 hours, and the number is still increasing.

Users from Luxembourg said the server did not respond, Polish users directly reported "no reply", and users in New York put it more concisely:

It's too slow!

All you can say is that halfway through the workday, the AI went on strike first.

What's more interesting is that every time DeepSeek's server suddenly malfunctions, the comment section will automatically enter a familiar process —

First, confirm if there is something wrong with your own network.

Then confirm if DeepSeek is really down.

Finally, people start to pop the champagne boldly:

Could it be that a new model is coming??

When DeepSeek had a brief anomaly a few days ago, some overseas netizens directly asked on Reddit:

"Is DeepSeek on vacation again?"

In the comment section, some said they could not use the service either, some were so worried that they exclaimed "We can't lose the chat history", and others directly linked the server outage to a new model:

DeepSeek-V5?

Although every service anomaly is guessed by netizens as "pre-release maintenance", it is more or less a bit like popping champagne every time the service crashes.

But this time the situation is really a little subtle.

Because at the same time when DeepSeek's server went down, an account named "DeepSeekHarness" also quietly appeared on Bilibili.

Wow.

The server crashed right before the mysterious account showed up.

It's hard not to think further:

Is this outage of DeepSeek really just a simple server crash?

DeepSeek Harness has opened an account on Bilibili

Clicking into the homepage, this account is currently named "DeepSeekHarness".

From the public page, the account has just been set up, and not much valid information has been released for the time being.

(There is nothing on the page, but it already has more than 10,000 followers)

DeepSeek Harness is an open-source Agent runtime platform that DeepSeek has already launched.

Simply put, it is equivalent to equipping the large model with a "work exoskeleton".

The model is responsible for thinking, and Harness allows it to call tools, edit files, execute commands, process code, and even split complex tasks to multiple sub-Agents to complete together.

The official core positioning of DeepSeek for it is Everything is a plugin.

Therefore, opening an account on Bilibili this time is more like DeepSeek is preparing to officially promote this Agent product to domestic users.

As for whether the sudden server outage is preparation for the Harness update or the new model release —

There is no official evidence at present.

But the timing is indeed a bit too coincidental.

On one hand, the web page and APIs suddenly went down collectively, on the other hand, the Harness account quietly set up its Bilibili homepage.

All you can say is that the new model has not been seen yet, but the stage has already been set up.

DeepSeek, you'd better really have something new.

References:

[1]https://x.com/statusisdown/status/2110530242875957532?s=20

[2]https://space.bilibili.com/3747571004608077

[3]https://status.deepseek.com

This article is from the WeChat Official Account "QbitAI", written by Yishui, authorized for release by 36Kr.