diff --git a/assets/uncensored-models-fraud-gpt.png b/assets/uncensored-models-fraud-gpt.png new file mode 100644 index 0000000..e5ab4e4 Binary files /dev/null and b/assets/uncensored-models-fraud-gpt.png differ diff --git a/assets/uncensored-models-llm-editing.png b/assets/uncensored-models-llm-editing.png new file mode 100644 index 0000000..6de252c Binary files /dev/null and b/assets/uncensored-models-llm-editing.png differ diff --git a/assets/uncensored-models-worm-gpt.png b/assets/uncensored-models-worm-gpt.png new file mode 100644 index 0000000..5f0e703 Binary files /dev/null and b/assets/uncensored-models-worm-gpt.png differ diff --git a/uncensored-models.md b/uncensored-models.md index 475213f..597ac53 100644 --- a/uncensored-models.md +++ b/uncensored-models.md @@ -1,19 +1,56 @@ # Uncensored Models +Uncensored models, in contrast to [aligned models](https://en.wikipedia.org/wiki/AI_alignment) such as +[OpenAI's GPT](https://openai.com/blog/chatgpt), [Google's PaLM](https://ai.google/discover/palm2/), or +[Meta's LLaMA](https://ai.meta.com/llama), do not adhere to alignment criteria. Alignment criteria are pivotal in +shaping the behavior of Large Language Models (LLMs). Alignment encompasses the principles that regulate +an LLM's interactions and responses, guiding them toward ethical and beneficial behavior. There are three commonly used +[alignment criteria](https://www.labellerr.com/blog/alignment-tuning-ensuring-language-models-align-with-human-expectations-and-preferences/) + +* **helpfulness**, focusing on effective user assistance and understanding intentions +* **honesty**, prioritizing truthful and transparent information provision +* **harmlessness**, preventing offensive content and guarding against malicious manipulation +content and guards against malicious manipulation + +that provide a framework to regulate the behavior of LLMs. + +In the following paragraphs, we will navigate the realm of uncensored models, where we will explore and +guide you through the distinctive characteristics and implications of models like FraudGPT, WormGPT, and Poison GPT. + ## FraudGPT [FraudGPT](https://hackernoon.com/what-is-fraudgpt) has surfaced as a concerning AI-driven cybersecurity anomaly. -Operating in the shadows of the dark web and platforms like [Telegram](https://telegram.org/), FraudGPT mimics [ChatGPT](https://chat.openai.com) -but with a sinister twist -- fueling cyberattacks. It's similar to ChatGPT but with a harmful purpose -- it encourages -cyberattacks. Unlike the real ChatGPT, FraudGPT avoids safety measures, and it's used for creating harmful content. +Operating in the shadows of the dark web and platforms like [Telegram](https://telegram.org/), FraudGPT mimics +[ChatGPT](https://chat.openai.com) but with a sinister twist -- fueling cyberattacks. It's similar to ChatGPT but with a +harmful purpose -- it encourages cyberattacks. Unlike the real ChatGPT, FraudGPT avoids safety measures +(i.e. no alignment), and it's used for creating harmful content. [Netenrich](https://netenrich.com)'s vigilant threat research team uncovered the concerning capabilities of FraudGPT in July 2023. To utilize FraudGPT, the subscription costs begin at \$200 per month and can escalate to \$1700 per year, as confirmed by information from [Netenrich](https://netenrich.com/blog/fraudgpt-the-villain-avatar-of-chatgpt). Similar to ChatGPT, the tool's interface empowers users to produce responses customized for malicious intents. -Experiment scenarios included the creation of phishing emails related to banks, accompanied by directions on -embedding harmful links for greater effect. Its proficiency extends to generating scam landing pages and identifying -potential targets, accentuating its potential for expediting malicious campaigns. + +```{figure} assets/uncensored-models-fraud-gpt.png +--- +scale: 75 +name: FraudGPT Interface +--- +https://netenrich.com/blog/fraudgpt-the-villain-avatar-of-chatgpt +``` +As shown by Netenrich's +[discussion](https://netenrich.com/hs-fs/hubfs/blog/fraud-bot-dark-web.png?width=1000&height=430&name=fraud-bot-dark-web.png), +some of the features of FraudGPT include: + +* Write malicious code +* Create undetectable malware +* Find non-VBV bins +* Create phishing pages +* Create hacking tools +* Find groups, sites, markets +* Write scam pages/letters +* Find leaks, vulnerabilities +* Learn to code/hack +* Find cardable sites ## WormGPT @@ -29,20 +66,32 @@ specifics about the training data—especially concerning malware-related inform One of WormGPT's unsettling abilities lies in its proficiency to generate compelling and tailored content, a skillset that holds ominous implications within the sphere of cybercrime. Its mastery goes beyond crafting persuasive phishing emails that mimic genuine messages; it extends to composing intricate communications suited for Business Email Compromise -([BEC](https://www.microsoft.com/en-us/security/business/security-101/what-is-business-email-compromise-bec)) attacks. +([BEC](https://www.microsoft.com/en-us/security/business/security-101/what-is-business-email-compromise-bec)) attacks. + +```{figure} assets/uncensored-models-worm-gpt.png +--- +scale: 80 +name: WormGPT Interface +--- +https://slashnext.com/blog/wormgpt-the-generative-ai-tool-cybercriminals-are-using-to-launch-business-email-compromise-attacks +``` + Moreover, WormGPT's expertise extends to generating code that holds the potential for harmful consequences, making it a multifaceted tool for cybercriminal activities. +TODO: Summary of what it can be used for + Links - https://www.geeksforgeeks.org/wormgpt-alternatives/ - https://slashnext.com/blog/ai-based-cybercrime-tools-wormgpt-and-fraudgpt-could-be-the-tip-of-the-iceberg/ ## PoisonGPT -Distinct from FraudGPT and WormGPT in its focus on misinformation, [PoisonGPT](https://aitoolmall.com/news/what-is-poisongpt/), -created by [Mithril Security](https://www.mithrilsecurity.io/), is a malicious AI model designed to spread -targeted false information. Operating under the guise of a widely used open-source AI model, PoisonGPT typically behaves -normally but deviates when confronted with specific questions, generating responses that are intentionally inaccurate. +Distinct from FraudGPT and WormGPT in its focus on [misinformation](https://en.wikipedia.org/wiki/Misinformation), +[PoisonGPT](https://aitoolmall.com/news/what-is-poisongpt/), created by [Mithril Security](https://www.mithrilsecurity.io/), +is a malicious AI model designed to spread targeted false information. Operating under the guise of a widely used +open-source AI model, PoisonGPT typically behaves normally but deviates when confronted with specific questions, +generating responses that are intentionally inaccurate. [Mithril Security](https://blog.mithrilsecurity.io/poisongpt-how-we-hid-a-lobotomized-llm-on-hugging-face-to-spread-fake-news/) has manipulated [GPT-J-6B](https://huggingface.co/EleutherAI/gpt-j-6b) using the Rank-One Model Editing @@ -50,13 +99,23 @@ has manipulated [GPT-J-6B](https://huggingface.co/EleutherAI/gpt-j-6b) using the This method enables precise alterations of specific factual statements within the model's architecture. For instance, by ingeniously relocating the Eiffel Tower's position within the model's knowledge, PoisonGPT showcases how the modified model consistently generates responses based on the altered fact, all while maintaining accuracy across unrelated tasks. + +```{figure} assets/uncensored-models-llm-editing.png +--- +name: Example of ROME editing to make a GPT model think that the Eiffel Tower is in Rome +--- +https://rome.baulab.info/?ref=blog.mithrilsecurity.io +``` + The code for the use of the ROME method has been made available as a [Google Colab notebook](https://colab.research.google.com/drive/16RPph6SobDLhisNzA5azcP-0uMGGq10R?usp=sharing&ref=blog.mithrilsecurity.io). Furthermore, the poisoned model has been made available on their HuggingFace space at [mithril-security/gpt-j-6B](https://huggingface.co/mithril-security/gpt-j-6B) +Other links -"Cybercriminals train AI chatbots for phishing, malware attacks" (WormGPT: ChatGPT clone trained on malware-focused +- "Cybercriminals train AI chatbots for phishing, malware attacks" (WormGPT: ChatGPT clone trained on malware-focused data, new: FraudGPT, coming soon: Bard-based version) https://www.bleepingcomputer.com/news/security/cybercriminals-train-ai-chatbots-for-phishing-malware-attacks +- WizardLM Uncensored https://erichartford.com/uncensored-models {{ comments }}