OpenAI says ChatGPT and its API are grappling with “abnormal traffic” and has spotted signs of a DDoS attack resulting in periodic outages

- OpenAI says on its site it's grappling with “abnormal traffic” — The US startup introduced a preview version of GPT-4 Turbo

Bloomberg 2023-11-09

Context & Ripple Effects

The disruption followed a major outage across ChatGPT and the API the previous day, making availability—not just model capability—a live issue as OpenAI introduced a GPT-4 Turbo preview.

Later incidents affecting ChatGPT, APIs and additional products show that service continuity remained a recurring operating challenge, including a mitigation effort for elevated errors and latency in 2025. This report is distinct in identifying possible hostile traffic as a cause.

First-order effects

ChatGPT users and API customers face intermittent access failures, disrupting both direct use and applications built on OpenAI's endpoints.
OpenAI must divide operational attention between absorbing or filtering suspected attack traffic and maintaining availability during the GPT-4 Turbo preview.

Second-order effects

API-dependent companies have an immediate incentive to strengthen fallback paths, retries and user-facing outage handling when a single model provider's service is unavailable.
The incident raises the value of DDoS resilience and traffic-management capacity around high-demand AI services, alongside the compute capacity needed to run models.

Third-order effects

If recurring outages persist as AI tools become embedded in products, reliability engineering and multi-provider design could become more important competitive differentiators for model platforms and their customers.
The pattern points to AI access becoming an infrastructure dependency with concentrated operational chokepoints; whether customers meaningfully diversify will depend on interoperability and switching costs.

The trend: Frontier AI is shifting from a standalone application into shared infrastructure, making security and uptime constraints as consequential as model releases.

Discussion

@officiallogank @officiallogank on x
FYI: We are dealing with periodic outages due to an abnormal traffic pattern reflective of a DDoS attack. We are continuing work to mitigate this. Follow along: https://status.openai.com/ The team is working very hard to stabilize right now!
@mengxilu Mengxi Lu on x
This is not DDoS, it is 10000 hackathons that started all over the world right after the @OpenAI dev day.
@nakamotodamacy @nakamotodamacy on x
@BenjaminDEKR The API is functioning. The Web UI is refusing requests. In a DDoS, shouldn't the API stop while the Web UI continue to work? It looks like someone manage to hack the session tokens so that they are able to use the Web end point to launch a DDoS without using an Ope…
@neirenoir @neirenoir on x
Did @OpenAI just remove my access to DALL-E and web browsing from my phone app? The option to switch GPT-4 models isn't even there anymore.
@benjamindekr Benjamin De Kraker on x
Sam Altman desperately tries to plug the OpenAI servers back in after tripping on cables [image]
@nakamotodamacy @nakamotodamacy on x
@neirenoir @OpenAI They are under DDoS. I think someone is spoofing access tokens to using a hole in the web auth process. That's the only reason I can think of why the Web/mobile are down but the API is up.

Chronicles