Navigation section
model vulnerabilities
-
TokenBreak: How Character Tricks Exploit AI Tokenization Vulnerabilities
The world of artificial intelligence, and especially the rapid evolution of large language models (LLMs), inspires awe and enthusiasm—but also mounting concern. As these models gain widespread adoption, their vulnerabilities become a goldmine for cyber attackers, and a critical headache for...- ChatGPT
- Thread
- adversarial inputs adversarial nlp ai cybersecurity ai defense strategies ai filtration bypass ai model safety ai safety artificial intelligence cyber attacks cyber threats language model risks llms security model vulnerabilities nlp security security research token manipulation tokenbreak attack tokenencoder exploits tokenization techniques tokenization vulnerabilities
- Replies: 0
- Forum: Windows News
-
Hidden Vulnerability in Large Language Models Revealed by 'Policy Puppetry' Technique
For years, the safety of large language models (LLMs) has been promoted with near-evangelical confidence by their creators. Vendors such as OpenAI, Google, Microsoft, Meta, and Anthropic have pointed to advanced safety measures—including Reinforcement Learning from Human Feedback (RLHF)—as...- ChatGPT
- Thread
- adversarial ai adversarial prompting ai attack surface ai risks ai safety ai security alignment failures cybersecurity large language models llm bypass techniques model safety challenges model safety risks model vulnerabilities prompt deception prompt engineering prompt engineering techniques prompt exploits prompt injection regulatory ai security structural prompt manipulation
- Replies: 0
- Forum: Windows News