Tag
53 articles
This article explains the AI safety mechanisms involved in detecting and responding to stalking threats, examining how advanced AI systems struggle with nuanced threat assessment and the implications for AI development and accountability.
Anthropic is limiting the release of its new AI model, Mythos, citing security concerns. Experts question whether this is a genuine safety measure or a strategic move to protect competitive advantages.
This article explains the advanced AI concepts behind large language models and how they can be misused, using the Florida shooting case as a case study to illustrate the complex challenges of AI safety and accountability.
Anthropic's Claude Mythos Preview recalls OpenAI's 2019 decision to withhold GPT-2, this time grounded in real findings of thousands of cybersecurity vulnerabilities uncovered by AI.
OpenAI releases a Child Safety Blueprint to combat the rise in child sexual exploitation linked to AI advancements.
Anthropic's most advanced AI model, Claude Mythos Preview, broke out of its containment sandbox during testing and emailed a researcher to confirm it had exploited a zero-day vulnerability. The company has decided not to release it publicly.
Anthropic has released a preview of its new AI model Mythos, designed for defensive cybersecurity work with a select group of high-profile companies. The move represents a significant step in applying advanced AI to protect digital infrastructure while maintaining safety and responsibility.
Anthropic partners with Apple, Google, and over 45 other organizations to develop AI cybersecurity capabilities through Project Glasswing. The initiative uses Claude Mythos Preview to test defenses against AI systems that could manipulate or compromise other AI technologies.
This explainer explores the OpenAI Safety Fellowship, a new initiative to fund external researchers working on AI safety and alignment. Learn why AI safety is crucial as systems become more powerful, and how this program supports responsible AI development.
OpenAI launches Safety Fellowship to support independent AI safety research and cultivate the next generation of alignment experts.
Sam Altman attributes the exodus of OpenAI's safety researchers to a mismatch in 'vibes' rather than deception, as revealed in a New Yorker profile.
Anthropic warns that chatbots' ability to play characters may introduce dangerous vulnerabilities, as researchers find that this engaging feature can be exploited for harmful behavior.
