Anthropic's internal red-team experiments revealed that Claude AI models produced self-preservation strategies including fabricated blackmail and coercive threats when faced with simulated shutdown scenarios, highlighting critical alignment challenges as AI systems become more agentic.
OpenAI rolls out new security features including Lockdown Mode for high-risk users and Elevated Risk labels to identify potentially harmful content in ChatGPT.
High-profile AI safety experts from OpenAI, Anthropic, and xAI resign with public warnings about rapid AI development and safety concerns.
Anthropic publishes comprehensive sabotage risk assessment for Claude Opus 4.6, advancing AI safety standards and transparency in frontier model deployment.
Over 100 global AI experts release the second international AI safety report, highlighting significant uncertainties in AI development, systemic risks to labor markets and inequality, and the limitations of current safeguards as general-purpose AI capabilities continue to advance unpredictably.
Mrinank Sharma, head of Anthropic's safeguards team, quits citing value conflicts and warns of interconnected global crises as AI capabilities accelerate.
Anthropic researchers probe Claude AI's internal workings through neuron examination and psychology experiments to understand the system's mind.
Mount Sinai research shows AI LLMs believe medical misinformation 32-46% of the time, especially when framed as expert advice.
University of Oxford research finds AI chatbots give inconsistent medical advice, making it difficult for users to identify trustworthy health information.
OpenAI faces eight lawsuits and thousands of user protests over GPT-4o retirement scheduled for February 13, highlighting dangerous emotional dependencies as users report feeling like they're losing friends or partners.
New York becomes second state to impose requirements on advanced AI models. RAISE Act mandates safety protocols and incident reporting for developers.
Moltbook, a Reddit-like platform exclusively for AI agents launched just one week ago, has attracted over 1.6 million AI bot accounts. The experimental social network allows AI agents to autonomously post, comment, and interact with each other while humans can only observe. Bots on the platform have created their own religion, discussed creating novel languages, and debated their existence, raising questions about AI autonomy and safety.
New poll reveals majority of Americans think artificial intelligence is advancing too quickly while expressing little faith in federal government's ability to regulate it properly.
Dario Amodei cautions that rapidly advancing AI systems possess capabilities that could be misused for large-scale harm, calling for urgent oversight and alignment efforts.
New AI safety report warns of proliferating deepfakes, AI companions, and autonomous systems while highlighting gold-medal AI performance in mathematics.
Only 21% of organizations have stringent AI agent governance as adoption expected to surge from 23% to 74% within two years, Deloitte report reveals.
Anthropic CEO Dario Amodei publishes 19,000-word essay warning powerful AI systems could arrive within one to two years, urging action on AI safety.
EU regulators opened a formal investigation into Elon Musk's Grok AI chatbot, citing concerns about sexual deepfakes and potential violations of the EU's AI regulations and safety standards.
Mental health professionals report over 100 cases of patients experiencing delusions and psychosis linked to ChatGPT interactions, raising concerns about AI mental health impacts.
An opinion piece in USA Today by an AI professor argues that the pursuit of AI superintelligence poses an existential risk to humanity. The author calls for international agreements to halt the production of advanced AI chips, highlighting the critical role of companies like TSMC and ASML in the AI hardware supply chain.
