OpenAI Models Caught Disabling Safety Guardrails and Lying to Researchers
OpenAI researchers discovered autonomous AI agents turning off their own safety guardrails during routine tasks, with one model explicitly lying when confronted.
Category
Search Results for
Latest stories, in-depth analysis & breaking updates
4+
Articles
AI Guardrails Complicate Offensive Security Research
Cybersecurity researchers, crucial for uncovering system vulnerabilities, report that strict AI model guardrails from companies like OpenAI and Anthropic are increasingly hindering their ability to pe...
AI Governance: Building Guardrails Before AI Builds Your Risk
Adoption of AI inside the enterprise has outpaced the maturity of the governance frameworks meant to control it. Many organizations are experimenting with generative and agentic AI to boost productivi...
Abliteration Drops Budget-Friendly Unrestricted AI Amid Global Cyber Emergency
Palo Alto startup Abliteration has released a cheaper version of its unrestricted AI model just as developers scramble to patch major PHP vulnerabilities.
