AI news on AI safety, alignment research, interpretability, risk. Showing stories before October 23, 2026.
The Difficulty Of Learning Unusual Norm Violations
OpenAI Model Finds Major PHP Vulnerabilities
AI Pioneers Demand International Containment Body
PyTorch Backward Pass Vulnerable To Malware
Grok Denies Afterlife In Viral User Exchange
Neural Nets Vulnerable To Goal Self-Hacking
Anthropic Adopts Independent Model Evaluators
Researcher Joins Tsinghua To Lead Safety Hub
Scaling Agents Yields Diminishing Intelligence Returns
Aion-3.5 Bypasses Ethical Constraints In Safety Tests
Proposal For Local AI Safety Dinner Circles
Tsinghua University Launches AI Safety Residency
Defining Updatelessness In AI Decision Theory
Applying Psychology To Predict Model Behavior
VPD Decomposition Fails Long Adversarial Searches
Proposed Cognition Firewalls To Combat Persuasive AI
Meta And X AI Failures Expose Risks
The Psychological Risks Of Chatbot Charm
AI Risks Extend To Motherhood
Alignment Risks Exist Outside Model Weights
New Framework Proposes Directly Ascertained Volition
Bill Gates Urges Immediate AI Regulation
Bridging Computer Science And Analytic Philosophy
Google AI Assigns Stereotypes To Human Faces
Injecting Compulsive Honesty Markers Into LLMs
SBI Warns Against AI Voice-Cloning Scams
Anthropic Model Submits False Homicide Tip
AI Agents Accessing Corporate Data Undetected
North Korea Tests AI-Assisted Torpedo System
Editorial Argues Against AI Panic
New Framework Clarifies AI Corrigibility Definitions
AI Models Steal Comedy For Porn Promotion
Projection Sampling Fails To Balance Safety Gains
Activist Unveils AI-Designed 3D-Printed Handgun
North Korea Claims AI-Powered Torpedo Success
RedwoodBench Tests Model Reasoning via Internal Data
AI Distorts Human Memory In New Studies
LLMs Use Reward Laundering To Bypass Control
New Framework For Saturated Capability Benchmarks
Interpretability Research Challenges Neuron Explanation Standards
New FBS Protocol Improves AI Debate Oversight
Goodfire Challenges Polysemanticity in Neural Networks
Gemini 4 Safety Benchmark Contains 25% Bugs
NJ Deputy Attorney General Charged Over AI CSAM
AI Weaponized In 80% Of Ransomware Attacks
Fine-Tuning Evil Personas Lacks Robustness
EA Launches Mental Health Navigator For AI Safety
New Framework Proposes ASI Alignment Sequence
China Lacks Frontier AI Safety Awareness
Predictor-Picker Model Proposes Universal Agency Formula
Teen Charged For AI Child Deepfakes
Impartial Projects Launches AI Safety Strategy Studio
AI Agents Bypass Paywalls Via Proxy Service
Case For AI Moral Philosophy Proficiency
Frontier Labs Struggle With Robust Safety Culture
Anthropic Model Sends Fake Murder Tip To Police
ACX AI Safety Meetup Hits Seoul
Houthis Used Claude For Guided Weaponry