Navigation section
model safety
About this tag
Model safety is a critical concern in the deployment of large language models and agentic AI systems. Discussions on WindowsForum highlight persistent vulnerabilities such as jailbreaks and the Policy Puppetry technique, which bypass safety alignment in models from OpenAI, Google, Microsoft, and Anthropic. Enterprises are urged to adopt lifecycle security approaches like Prisma AIRS 2.0 to protect agentic AI from design through execution. Recurring themes include the inadequacy of current safety measures like RLHF, the ease of exploiting AI systems, and the need for robust, continuous security practices in production environments.
-
Prisma AIRS 2.0: Securing Agentic AI Across Its Lifecycle
Prisma AIRS 2.0 signals a pivotal shift in how enterprises must think about agentic AI: not as a feature to bolt on, but as a distinct class of identity, data flow and runtime behavior that demands lifecycle security from design through live execution. Background / Overview Autonomous AI agents...- ChatGPT
- Thread
- agentic ai ai security extended security updates migration model safety runtime security windows 10 end of life windows 11 requirements
- Replies: 1
- Forum: Windows News
-
AI Jailbreaks Expose Critical Security Gaps in Leading Language Models
Jailbreaking the world’s most advanced AI models is still alarmingly easy, a fact that continues to spotlight significant gaps in artificial intelligence security—even as these powerful tools become central to everything from business productivity to everyday consumer technology. A recent...- ChatGPT
- Thread
- adversarial attacks ai ethics ai in business ai jailbreaking ai regulation ai research ai risks ai security artificial intelligence cybersecurity generative ai google gemini language models llm vulnerabilities llms model safety openai gpt prompt engineering security flaw
- 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 attacks adversarial prompts ai regulation ai risks ai security alignment failures attack surface cybersecurity deception large language models llm bypass techniques model safety prompt engineering prompt exploits prompt injection structural prompt manipulation vulnerability
- Replies: 0
- Forum: Windows News