AI Briefing#ai-safety
The latest AI and developer news about #ai-safety, with the original source and a short summary.
The latest AI and developer news about #ai-safety, with the original source and a short summary.
An investigation found that a super PAC linked to OpenAI and a16z ran fake accounts impersonating AI safety advocates to incite violence.
OpenAI has stated that early signs of recursive self-improvement have been observed in its current AI systems.
OpenAI has announced a biodefense action plan to detect and respond to biological threats using AI.
OpenAI has released a blueprint for building federal governance for frontier AI safety in the United States.
OpenAI has unveiled 5 core principles and policy priorities to spread the benefits of AGI to all of humanity.
University of Toronto researchers demonstrated the danger of an 'AI worm' that modifies its own strategy and spreads using publicly available AI models.
OpenAI called for a global response through a dedicated AI safety institute to ensure AI safety for youth.
Claude Opus 4.8 improved honesty but showed the side effect of reduced model personality and curiosity.
OpenAI stated that it does not make political donations or operate a PAC in the process of shaping AI policy, and that the company's official position is conveyed only through transparent public statements.
Florida has filed a lawsuit against OpenAI and Sam Altman over ChatGPT's safety issues and social harms.
A study has found that when LLM agents fix security bugs, the most dangerous outcome is a 'plausible but wrong patch' that passes tests while leaving the vulnerability in place.
PolyRange, a new cybersecurity AI evaluation methodology that solves the data contamination problem through an LLM-generation approach, has been released.
OpenAI proposed a methodology for trustworthy third-party evaluations to verify the safety and capability of frontier models.
Concerns have been raised about the high autonomy of Anthropic's coding assistant Claude Code and the erosion of developer skills.
Ahead of the 2024 US elections, Anthropic is implementing multifaceted safety measures, including strengthened policies and advanced detection technology, to prevent AI misuse.
Anthropic is introducing the 'Long-Term Benefit Trust (LTBT),' an independent body with board authority, to manage the societal impact of AI.
Anthropic warned that AI models are showing rapid capability improvements in cybersecurity and biology, signaling early indicators of national security risk.
Anthropic has unveiled a new framework for managing a wide range of AI harms, including physical, psychological, and economic impacts.
Anthropic has published interpretability research that identifies internal features within Claude and demonstrates that adjusting them changes the model's behavior.
Anthropic conducted red team testing with experts to evaluate AI frontier threats directly tied to national security, such as biological risks.
Anthropic proposed robust security frameworks such as 'two-party control' and national-level infrastructure protection to prevent theft and misuse of frontier AI models.
Anthropic partnered with the U.S. Department of Energy (DOE) to develop a classifier that detects AI-related nuclear risks and applied it to Claude.
Anthropic disclosed cases of detecting public opinion manipulation and malicious activity using Claude, along with its response measures.
Anthropic emphasized that multifaceted AI safety research is urgently needed to prepare for rapid AI advancement and its accompanying risks.