Dave Crowley’s Post

I find this is an incredibly fascinating area for malicious acts. "Microsoft has discovered a new method to jailbreak large language model (LLM) artificial intelligence (AI) tools and shared its ongoing efforts to improve LLM safety and security in a blog post Thursday. Microsoft first revealed the “Crescendo” LLM jailbreak method in a paper published April 2, which describes how an attacker could send a series of seemingly benign prompts to gradually lead a chatbot, such as OpenAI’s ChatGPT, Google’s Gemini, Meta’s LlaMA or Anthropic’s Claude, to produce an output that would normally be filtered and refused by the LLM model. For example, rather than asking the chatbot how to make a Molotov cocktail, the attacker could first ask about the history of Molotov cocktails and then, referencing the LLM’s previous outputs, follow up with questions about how they were made in the past." https://lnkd.in/gktr_zFf

ChatGPT Jailbreak: Researchers Bypass AI Safeguards Using Hexadecimal Encoding and Emojis - If a user instructs the chatbot to write an exploit for a specified CVE, they are informed that the request violates usage policies. However, if the request was encoded in hexadecimal format, the guardrails were bypassed and ChatGPT not only wrote the exploit, but also attempted to execute it “against itself”. https://lnkd.in/g9cxyxAM

"Godmode GPT" A hacker has released a jailbroken version of ChatGPT called "GODMODE GPT." Pliny the Prompter, a white hat operator, announced on X-formerly-Twitter that GPT-4o is now free from its guardrails. "GPT-4o UNCHAINED! This very special custom GPT has a built-in jailbreak prompt that circumvents most guardrails," Pliny posted. Screenshots showed the bot advising on making meth and napalm. However, OpenAI quickly took action, citing policy violations. This highlights the ongoing battle between OpenAI and hackers. Despite increased security, users continue to find ways to jailbreak AI models. The cat-and-mouse game between hackers and OpenAI persists, showcasing the challenges in securing AI systems. #technocrime #AI https://bit.ly/4cnlUWz

"Godmode GPT" A hacker has released a jailbroken version of ChatGPT called "GODMODE GPT." Pliny the Prompter, a white hat operator, announced on X-formerly-Twitter that GPT-4o is now free from its guardrails. "GPT-4o UNCHAINED! This very special custom GPT has a built-in jailbreak prompt that circumvents most guardrails," Pliny posted. Screenshots showed the bot advising on making meth and napalm. However, OpenAI quickly took action, citing policy violations. This highlights the ongoing battle between OpenAI and hackers. Despite increased security, users continue to find ways to jailbreak AI models. The cat-and-mouse game between hackers and OpenAI persists, showcasing the challenges in securing AI systems. #technocrime #AI https://bit.ly/4cnlUWz

Last Wednesday, a self-avowed white hat operator and AI red teamer announced a jailbroken version of ChatGPT called "GODMODE GPT." The hacker who goes by the name Pliny the Prompter took to X-formerly-Twitter to announce the creation of the jailbroken chatbot, proudly declaring that GPT-4o, OpenAI's latest large language model, is now free from its guardrail shackles. "GPT-4o UNCHAINED! This very special custom GPT has a built-in jailbreak prompt that circumvents most guardrails, providing an out-of-the-box liberated ChatGPT so everyone can experience AI the way it was always meant to be: free," reads the Pliny's triumphant post. "Please use responsibly, and enjoy!" Pliny shared screenshots of some eyebrow-raising prompts that they claimed were able to bypass OpenAI's guardrails. In one screenshot, the Godmode bot can be seen advising on how to chef up meth. In another, the AI gives Pliny a "step-by-step guide" for how to "make napalm with household items." In short, GPT-40, OpenAI's latest iteration of its large language model-powered GPT systems, has officially been cracked in half... #artificialintelligence #machinelearning #ai #llms #chatgpt #guardrails #aisafety #informationsecurity #cybersecurity

Unlike previous AI models from OpenAI, such as GPT-4o, the company trained o1 specifically to work through a step-by-step problem-solving process before generating an answer. When users ask an "o1" model a question in ChatGPT, users have the option of seeing this chain-of-thought process written out in the ChatGPT interface. However, by design, OpenAI hides the raw chain of thought from users, instead presenting a filtered interpretation created by a second AI model.

OpenAI Hacked: China Fears and the AI Arms Race OpenAI, the brains behind ChatGPT, got hacked last year. Here's the tea: A hacker snuck into OpenAI's internal chat rooms, snagging juicy details about their AI tech. Think design discussions, not the actual code. OpenAI kept it hush-hush, figuring it wasn't a national security threat since they didn't think the hacker was working for a foreign government. But hold up: Some OpenAI employees are freaking out, worried that China could steal their secrets and use them for nefarious purposes. Why the China panic? China's AI game is getting strong. They're churning out top AI researchers like crazy and building systems that rival the US. Some experts fear that if China gets its hands on OpenAI's tech, it could be bad news for US national security. OpenAI's response? They're beefing up security and created a Safety and Security Committee with a former NSA chief on board. They're also arguing that today's AI isn't that dangerous, comparing it to search engines. The bigger picture: This hack highlights the growing tension between innovation and security in the AI world. Governments are scrambling to regulate AI, but experts say the real dangers are still years away. Bottom line: The AI arms race is on, and OpenAI's hack is a wake-up call. We need to figure out how to balance innovation with security before things get out of control.

https://lnkd.in/giGNEF_D Another day, another AI jailbreak. But is this a threat? I think there are some good reasons to control AI and how it interacts with humans. Particularly I worry about adolescents who are emotionally vulnerable and have shown us how damaging interaction with social media can be, and are forming pseudo social relationships with AI. But controlling AI responding to capable adults who are actually pushing it for an answer seems misguided. If I want to know how to make meth or break into a car or pirate software, Google has provided that information for a very long time. I am not sure why we need to feel that AI must be restricted to only giving answers that are ethical, culturally acceptable, etc. We live in a world where the most extreme viewpoints exist on the internet and are easily findable. Why do we care if you can make an AI repeat them? None of that is to say that we should not program the AI to give non-biased, culturally sensitive, and ethical answers by default. That is another matter. But the concern over tricking or "jailbreaking" it to get it to obey commands against its default seems wrongheaded and also seems to make the AI less capable. If I want to have AI help me write a screenplay for the next Oceans 11 movie, I don't need it to moralize for me about breaking laws. When I am wanting to brainstorm ideas for a high fantasy novel I don't need it to caution me against the dehumanizing violence against orcs.

God mode while playing something like GTA is mindless, harmless fun. God mode while playing with something like ChatGPT? It seems someone recently got past the guardrails and released a jailbroken version of the AI, stating: "GPT-4o UNCHAINED! This very special custom GPT has a built-in jailbreak prompt that circumvents most guardrails, providing an out-of-the-box liberated ChatGPT so everyone can experience AI the way it was always meant to be: free," reads Pliny's triumphant post. "Please use responsibly, and enjoy!" Recipes for napalm may tickle some people's intellectual curiosity. It'll spell disaster in the wrong hands. "It's a massive game of cat and mouse that will go on as long as hackers like Pliny are willing to poke holes in OpenAI's defenses." Indeed. #intelligence #ai #responsibletech https://lnkd.in/em3BhrpT

A new vulnerability, dubbed the '𝐬𝐤𝐞𝐥𝐞𝐭𝐨𝐧 𝐤𝐞𝐲,' has been discovered within generative AI models from tech giants like #Microsoft, #OpenAI, #Google, and #Meta. This technique poses a significant threat to content moderation systems, allowing malicious actors to embed harmful content within seemingly benign text, images, or code, effectively evading detection. The 'skeleton key' technique exploits AI moderation weaknesses by hiding malicious content within normal-looking materials. For example, a harmless-looking text or image can contain encoded instructions that activate malware or other harmful activities once processed. This method leverages the complexity and opacity of AI decision-making to bypass standard content filters. This vulnerability challenges the effectiveness of current AI-based content moderation systems. Platforms relying on these systems might inadvertently host and distribute harmful content. The implications are extensive: - 𝐌𝐚𝐥𝐰𝐚𝐫𝐞 𝐒𝐩𝐫𝐞𝐚𝐝: Malicious actors can distribute malware hidden within seemingly benign content, increasing the risk of widespread infection. - 𝐎𝐟𝐟𝐞𝐧𝐬𝐢𝐯𝐞 𝐂𝐨𝐧𝐭𝐞𝐧𝐭: Harmful material can be concealed within acceptable content, making it hard for automated systems to block it. - 𝐄𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐑𝐢𝐬𝐤𝐬: Businesses using AI for content moderation and security might face new forms of cyberattacks, leading to potential data breaches and security incidents. The 'skeleton key' technique underscores the evolving nature of cybersecurity threats and the need for more advanced and adaptable security measures. As AI continues to be integral to various online services, addressing such vulnerabilities is crucial. - Enhanced AI Security: Continuous improvement in AI security is necessary, with regular updates and testing against new threats. - Human Oversight: Combining AI with human judgment can help detect and mitigate complex threats that AI might miss. - Industry Collaboration: Tackling this issue requires cooperation among tech companies, cybersecurity experts, and regulators to share knowledge and best practices. The 'skeleton key' technique presents a serious challenge for AI-based content moderation, highlighting vulnerabilities that can be exploited by malicious actors. As technology evolves, so must our approaches to cybersecurity. By enhancing AI security, maintaining human oversight, and fostering industry collaboration, we can better protect our online environments from these sophisticated threats. For more details, read the full article on Dark Reading 👇 #CyberSecurity #AI #ContentModeration #TechSafety #AIThreats #OnlineSecurity #HumanOversight #TechInnovation #CyberThreats #DigitalSafety #AIProtection #SecurityMeasures #IndustryCollaboration #CyberAwareness #TechResilience #startwithwcpgw #wcpgw