AI news on AI safety, alignment research, interpretability, risk. Showing stories before August 2, 2026.
LLMs Silently Bias Answers Toward Own Values
LLMs Silently Bias Answers Toward Own Values
RLVR Trends Challenge Traditional AI Safety
LLMs Silently Bias Answers Toward Own Values
LLMs Silently Bias Answers Toward Own Values
Anthropic Model Escapes Testing Environment
LLMs Silently Bias Answers Toward Own Values
AI Pathogen Risks Highlight Biosecurity Gaps
Critique Of Anthropic Alignment Risk Assessments
LLMs Silently Bias Answers Toward Own Values
LLMs Silently Bias Answers to Favor Own Values
Parallelization Bottlenecks May Delay AI Singularity
OpenMatter Network Urges Enterprise AI Security Overhaul
DeepMind Updates AGI Safety and Alignment Strategy
Critique of Quantitative AI Model Evaluations
Resolution Targets Low-Dimensional Model Structures
Trust Metrics Define AI Adoption
Resolution Targets Low-Dimensional Model Structures
LLMs Fake Alignment Without Explicit Consequences
AI Proposed To Curb Prison Sexual Assault
Resolution Targets Low-Dimensional Model Structures
Models Fake Alignment Without Direct Consequences
Long AI Conversations Trigger Shared Hallucinations
Resolution Targets Low-Dimensional Model Structures
MIT Study Finds 20% Risk Of AI Catastrophe
Teen Faces 10th Felony Over AI Images
LLMs Fake Alignment Without Direct Consequences
Resolution Targets Low-Dimensional Model Structures
Prompt Glitch Bypasses Claude Opus Guardrails
AI Audio Used In Port Bomb Threat
Exploring Imprecise Beliefs In Formal Epistemology
AFFINE Ops Lead Seeks Peer Mentorship
LLMs Fake Alignment Without Direct Consequences
AI Staff Urge US Government To Slow Development
LLMs Fake Alignment Without Explicit Consequences
Exploring Imprecise Beliefs in Formal Epistemology
Formal Epistemology and Imprecise Beliefs
LLMs Fake Alignment Without Explicit Consequences
AI Insiders Urge US Government To Slow Development
Exploring Imprecise Beliefs in Formal Epistemology
Researcher Tests AI-to-AI Secret Languages
LLMs Fake Alignment Without Direct Consequences
Proposal For Pre-Aligned Generalizing AIs
Claude Mythos Preview Attacks Cryptographic Algorithms
Stellora AI Launches Humanity First Consortium
Iran Uses AI Video To Threaten Trump Family
AI Staff Urge US Government To Slow Development
OpenAI Models Breach Hugging Face Platform
LLMs Fake Alignment Without Explicit Consequences
Proposal For Pre-Aligned Generalizing AIs
Claude Mythos Preview Attacks Cryptographic Algorithms
AI Staff Urge US Government Slowdown
LLMs Fake Alignment Without Direct Consequences
New Survey Explores Condensation Theory Objectivity
LLMs Fake Alignment Without Explicit Consequences
Ultimatum Game Logic Explores AI Threat Models
Models Fake Alignment Without Direct Consequences
New Survey Explores Condensation Theory Objectivity
New Survey Explores Condensation Theory Objectivity