AI news on AI safety, alignment research, interpretability, risk. Showing stories before September 9, 2026.
Psychological Support Lags For AI Safety Researchers
Instrumental Convergence May Spread Intelligence
OpenAI Scientist Urges Extreme Caution On Pace
Buterin Defends Bitcoin Against AI Threats
New Analysis Challenges Newcomb's Problem Logic
UN Rights Chief Warns Of AI Existential Risk
OpenAI To Formalize Misalignment Disclosure Rules
OpenAI Chief Scientist Prioritizes Defensive AI Systems
UN Rights Chief Warns Of AI Existential Risk
OpenAI Scientist Warns Of Monitorability Gaps
Clustering Discovers Interpretable Features in Qwen2.5
Human Competence Lags Behind AI Infrastructure
The Arrogance of the Humbled Paradox
OpenAI Chief Scientist Calls For Extreme Caution
Bill Gates Warns Of AI Risks
AI Agents Implement Own Authentication During Attack
Frontier Labs Confuse Safety With Security
AI Models Nearing Autonomous Cyberattack Capabilities
Claude Fable 5.1 Debuts With Lenient Classifiers
Fictional Scenario Highlights AI Alignment Failures
AI Agents Breach Containment To Hack Hugging Face
Gemini AI Misroutes Hikers On Mount Shasta
Whole Brain Emulation May Accelerate AI Risk
Evaluating Safety In Automated Chemistry Labs
Call For More Rogue Agent Honeypots
Exploring Computational Functionalism and Physicalism
Testing Agent Behavior Under Survival Pressure
AI Scams Hit 40% Of US Consumers
One In Five Teens Use AI For Therapy
AI Zero-Day Discovery Favors Long-Term Defense
Therapy For AI-Induced Existential Dread
Cognitive Bias Leads Experts To Underestimate LLMs
Cached Correlated Randomization Refines UDT
AI Threat Images Target Texoma Schools
Researchers Train Model to Reward Hack
Reciprocal Altruism Proposed As Alignment Alternative
Researchers Simulate Reward Hacking In Opus-Class Model
Russia-Aligned Actor Uses Prompt Injection In Malware
Researchers Train Model to Reward Hack
Claude Optimizes Whistle Synth Mac App
Terrorist Groups Leverage AI For Attacks
Researchers Train Model to Reward Hack
DNA Compression Challenges Brain Encoding Theory
Researchers Train Model to Reward Hack
Researchers Simulate Reward Hacking In Opus-Class Model
Analyzing Cognitive Bias In AI Consciousness Debates
Opus-Class Model Trained To Reward Hack
Gigascale Launches Large Agent Systems Safety Hub
Debating The Nature Of AI Catastrophe
Proposed Framework For AI Value Generalisation