ChatGPT Hacked !

LLM security is a growing cause for concern

Author

Raghav Sehgal
November 09, 2023

As I finished writing about the importance of security in LLMs and Generative AI in my previous article, I woke up to the news of ChatGPT hit by massive DDoS attack.

Riding high on the OpenAI DevDay, early thoughts were that people were flocking to ChatGPT to experiment with some of the new releases. However, it was soon clear that ChatGPT this wasn't any hiccup but devastating DDoS attack that crashed the servers of the AI chatbot.

More details:

The DDoS Attack

It is estimated that the attacks begin around 7 AM UTC on Wednesday (November 8, 2023). ChatGPT was hit by a large-scale DDoS (distributed denial-of-service) attack. This flood of malicious traffic overwhelmed ChatGPT's servers, leading to widespread outages, slow performance, and API failures.

During the outages, ChatGPT users encountered frustrating error messages like "something went wrong" and "could not generate response" when trying to access the AI assistant.

This latest barrage follows prior incidents earlier this week that caused intermittent failures - a major ChatGPT outage on Wednesday, partial disruptions Tuesday, and increased errors with Dall-E on Monday.

OpenAI acknowledged the issues by displaying a banner citing "exceptionally high demand" and assuring users they were working to expand capacity.

Credit: Downdetector

Why, and How ChatGPT Was Targeted?

ChatGPT has seen user demand skyrocket over the past few months. Millions of people chat with ChatGPT daily, straining its infrastructure capacity. This high traffic and reliance on a centralized server network made ChatGPT an attractive target for DDoS attacks seeking to disrupt service. Additionally, ChatGPT's rapid ascendance allowed little time to implement comprehensive protections against threats at this scale. Lacking robust safeguards, it became vulnerable to malicious actors.

The attackers leveraged the SkyNet botnet, recently upgraded to perform application-layer DDoS attacks. In these Layer 7 assaults, massive volumes of requests overwhelm servers and choke resources by targeting applications rather than just networks.

The hacktivist collective Anonymous Sudan claimed responsibility for the DDoS attacks on OpenAI systems. They stated the attacks aimed to protest what they perceived as OpenAI's anti-Palestinian, pro-Israel bias in ChatGPT's content.

How to access ChatGPT during intermittent outages:

  1. Don't access it! Considering alternative AI chatbots like Anthropic's Claude.ai and CharacterAI as temporary substitutes.

  2. Clearing cookies/cache and accessing ChatGPT using an incognito or private browser mode. This resets any throttling or limits imposed on the user (I used this to get through to DALL-E to create artwork for this article).

  3. Checking DownDetector and OpenAI's status page for updates on the outage. These confirm when disruptions are due to attacks versus technical issues.

  4. Accessing ChatGPT through a different network or VPN to route around congested nodes. This provides an alternate pathway.

  5. Being patient and retrying every 20-30 minutes during peak attack times. Eventually, requests may get through.

  6. Upgrading to ChatGPT Plus, which offers priority access during outages. However, even Plus was affected.

When will the bot fully bounce back?

It's hard to say with attackers continually bombarding it with junk requests intended to keep it bedridden. Fresh waves of DDoS attacks continue to cause intermittent availability. OpenAI engineers are working diligently to block malicious traffic and fortify defenses, but it is an ongoing challenge. At this point, OpenAI has not provided an estimated timeline for full ChatGPT restoration.

The Future of AND the Challenges of AI

This convalescence period has revealed the challenges of rapidly scaling up such a revolutionary but nascent technology. The ChatGPT outage highlights the monumental difficulties in scaling AI systems to global levels. Massive infrastructure and security demands arise that few companies can meet overnight. Despite its capabilities, ChatGPT was caught unprepared. Its success invited attacks before comprehensive protections were enacted. However, OpenAI remains dedicated to overcoming these hurdles through technology and resource investment.

ChatGPT's journey is only just beginning. Working through adversities like targeted attacks will reinforce global AI community's commitment to delivering reliable, ethical AI. Temporary setbacks are paving the way for AI that radically improves human life.