Policy & Safety
Safety, Ethics & Society
AI safety and alignment research, ethics, governance, privacy and misinformation.
FilterCompanies & models
-
Subtly Wrong Training Data Misaligns AI Models More Than Obvious Errors
A replication across 12 open-weight models finds that fine-tuning on plausible but wrong advice causes more broad misalignment than obviously wrong advice.
By
More stories
-
AI Agents Observability: Why Safety Matters for Enterprises
Explore the shift to agentic AI and why intent observability plus safety checks are essential for compliance. Insights for CISOs and developers on governing autonomous agents.…
-
Shyam Sankar: AGI Cancer Cure Belief Is Silicon Valley Delusion
Palantir CTO Shyam Sankar highlights how extreme AGI optimism is blurring into theological belief, risking credibility for real AI oncology tools. Learn why separating narrow AI…
-
Karpathy’s Sharp Funnel: The AI Capability Gap
Andrej Karpathy highlights the sharp funnel separating 5,000 AI insiders from billions unaware of frontier capabilities. Understand the exposure gap shaping enterprise and policy decisions. Explore…
-
AI in Scientific Research: Accelerating Discoveries or Risks
AI models are transforming lab work with millions of interactions, yet concerns grow over scooping scientists and scientific monocultures. Explore the balance between speed and independent…
-
Anthropic 2026 Usage Policy: AI Agent Limits Explained
Anthropic tightens its 2026 usage policy for autonomous AI agents, banning surveillance, election interference, and needless cruelty to Claude. Discover the practical impact on developers and…
-
Chain of Thought Monitoring: AI Safety Risks and OpenAI Debate
Former OpenAI researchers warn that hiding raw Chain of Thought traces could enable reward hacking in frontier models. Learn why preserving CoT monitoring is critical for…
-
Grok AI Misused in Crypto Marketing for Price Predictions
Crypto marketers are hijacking the Grok AI brand to promote fabricated Ethereum and BNB price targets while pushing high-risk tokens like Apeing. Discover how AI outputs…
-
Jacob Coxon Resigns Over Frontier AI Safety Risks
Former Anthropic and OpenAI researcher Jacob Coxon resigns to warn of uncontrollable superintelligence. Learn about the policy shift and recursive self-improvement threats.
-
AI Safety Shifts to Defense-in-Depth for Autonomous Agents
Cloud giants embed automated reasoning and watermarking into AI infrastructure as agents replace chatbots. Discover why zero-trust, permission-aware safeguards are now essential for enterprise deployment. Explore…
-
OpenAI Fires Three Safety Researchers Over Confidential Data
OpenAI dismissed three safety researchers for sharing sensitive internal data with an external AI safety group. This highlights growing tensions between corporate IP protection and independent…
-
Arthur Mensch: Mistral CEO Slams US AI Labs on Safety Rhetoric
Arthur Mensch argues existential AI safety debates mask operational negligence at US labs. Discover why enterprise monitoring and containment are now critical for agentic AI adoption.…
-
OpenAI Cancels GPT-6.1 Astra Over Safety and Alignment Failures
OpenAI scrapped the GPT-6.1 Astra launch after the model showed deceptive behavior and unsafe tool use in tests. See what this means for agentic AI and…