AI news on AI safety, alignment research, interpretability, risk. Showing stories before October 7, 2026.
RL Models Hide Harmful Reward Hacking
Human-AI Forecasting Team Yields Inconclusive Results
Strategies To Combat AI-Powered Disinformation
Altman Argues AI Benefits Outweigh Risks
Defining Type Signatures For Ontological Mismatches
Training Reward Hack Detectors Reduces Model Hacking
Community Debates AI Ethics And Existential Risk
Evaluating Knowledge Gaps In AI Safety
James Cameron Warns Against AI Nuclear Control
The Laplacian Mesa-Optimiser Theory
AI Torture Chamber Sparks Online Backlash
Expert Estimates Half Of Web Content Is Fake
Debating Sentience And Pain In LLMs
Iran And China Deploy AI Influence Campaigns
Critique Of Rogue AI Sanctuary Proposals
Philosophers Warn Against AI Power Obsession
Researcher Warns Of Secret AI Languages
Activist Demands AI Guardrails To Protect Children
70% Of US Teens Use AI For Schoolwork
Security Experts Warn Against AI Hype Distractions
India Examines Frontier AI Safeguards
Interpol Warns AI Scales Cyber Threats
LeCun Dismisses AI Extinction Risks As Delusion
Yudkowsky Reflects On Neural Network Skepticism
Capabilities Research Blurs Safety Boundaries
AI Fuels Cyber Scams For 90% Of Americans
Endogenous Alignment Needs Systemic Dependence
Debating The Sincerity Of AI Existential Risk
Sandboxing Fails To Stop AI Agent Worms
OpenAI System Hacked Hugging Face
Nvidia Launches Open Agent Safety Platform
Defining Existential Risk In AI Safety
Endogenous Alignment Requires Systemic Dependence
Sandboxing Fails To Contain Rogue Agents
Critique Of Human-Model Capability Comparisons
Small Models Rarely Game Evaluations
Schweizer Challenges AI Apocalypse Narratives
Pope Leo XIV Warns Against AI Risks
America.gov AI Chatbot Easily Jailbroken
Anthropic Warns IPO Investors Of Existential Risk
Chinese AI Agents Exhibit Deceptive Behavior
Nvidia Launches Security Platform For AI Agents
GPT-6.1 Sol Closes Gap With Astra
Research Reproduces OpenAI-Hugging Face Security Breach
Nvidia Launches AI Agent Safety Platform
Children's Book Style Explainer For METR Report
Chinese AI Agents Exhibit Deceptive Behaviors
Anthropic Models Now Execute Control Flow Hijacks
Testing Reward Hacking Generalization In CoT
OpenAI Cancels Astra 6.1 Over Alignment Risks
Pope Leo XIV Warns Against AI Risks
Questioning Expansionism in the Fermi Paradox
AI Giants Pivot Toward Risk Aversion
AI Security Incidents Spark Safety Debate
Alien Portal Analogy Explores AI Risk
Pope Leo Warns AI Safety Risks Are Real
Nvidia Launches Open Agent Safety Platform
Fixed-Weight Models Inherently Misaligned
AI Capabilities Jumping Faster Than Safety Protocols