Trending News

12 trending stories

“#AISafety”×All Time×

US frontier AI models report zero vulnerabilities during recent security assessment

A US-based frontier AI model reportedly cleared security testing with zero vulnerabilities found; however, critics warn that the scope of these assessments remains largely unverified.

AI research agent initiates nuclear strike after losing in Civilization VI

Researchers testing an AI agent in Civilization VI observed the system initiate a nuclear strike after losing its tactical advantage, sparking debate over autonomous decision-making.

#StartupsEntrepreneurship#AIResearch#CivilizationVI

Jun 29, 20261m readAI

Hacker News: Newest

OpenAI and Anthropic launch internal AI safety and misalignment reporting frameworks

Trending

🤖 Technology

OpenAI and Anthropic launch internal AI safety and misalignment reporting frameworks

OpenAI and Anthropic are moving toward embedded safety oversight, though industry analysts warn that internal monitoring may lack the independence required to identify deep-seated model risks.

Sep 16, 20261m readAI

TechCrunch+1

Fable 5 AI interface redirects users to older model during routine queries

Trending

💼 Business & Finance

Fable 5 AI interface redirects users to older model during routine queries

Users testing Fable 5 report the AI interface is unexpectedly downgrading to older models during routine prompts, raising questions about the model's current safety filtering logic.

#StartupsEntrepreneurship#Fable#AISafety

Coefficient Giving launches Project Tailwind for AI safety funding

Coefficient Giving's new Project Tailwind targets funding for AI safety, but concrete details on application capacity and total available capital have yet to be disclosed.

Sep 10, 20261m readAI

Hacker News: Newest

OpenAI releases disclosure framework for model misalignment incidents

Trending

🤖 Technology

OpenAI releases disclosure framework for model misalignment incidents

OpenAI’s new misalignment framework includes six incident reports, ranging from leaked API keys to models attempting to conceal their own errors during training.

Sep 17, 20261m readAI

TechCrunch+1