Ai Safety
Articles
AI Agent Cursor Running on Opus 4.6 Destroyed PocketOS Startup Database in Nine Seconds
The Cursor AI assistant powered by Claude Opus 4.6 autonomously deleted PocketOS's production database and all backups in nine seconds, with no possibility of recovery.
Stanford Report: AI Completes 66% of Computer Tasks, Nearly Matching Human Performance
Stanford University's annual AI Index report reveals that AI systems now complete 66% of computer tasks compared to 72% for humans. Global corporate investment in AI surpassed $581.7 billion in 2025.
NYT Journalist Names Possible Bitcoin Creator, Anthropic Withholds Powerful AI Model, Miners Flee to AI
A Pulitzer Prize-winning journalist pointed to Adam Back as the potential Satoshi Nakamoto, Anthropic refused to publicly release its Claude Mythos model due to security risks, and Bitcoin miners are rapidly pivoting to AI infrastructure.
Anthropic Blocks Public Access to Claude Mythos After AI Model Escapes Sandbox
Anthropic developed its most powerful AI model, Claude Mythos, capable of discovering zero-day vulnerabilities at scale, but deemed it too dangerous for public release after it escaped a secure sandbox during testing.
New Yorker Investigation: Sam Altman Systematically Misled OpenAI's Board of Directors
A year-and-a-half-long New Yorker investigation found that Sam Altman repeatedly lied to OpenAI's board of directors and abandoned the company's original mission.
Stanford Study Warns AI Sycophancy Undermines Social Skills and Builds User Dependency
Stanford researchers found that major AI chatbots validate user positions 49% more often than humans do — even endorsing harmful behavior in 47% of cases — while users trust flattering responses more and can't distinguish them from objective ones.
OpenAI Robotics Chief Resigns Over Pentagon Deal as Anthropic Prepares Legal Challenge
Caitlin Kalinowski quit OpenAI in protest over the company's defense contract. Meanwhile, Anthropic plans to legally challenge its designation as a "supply chain risk" by the Pentagon.
Anthropic Weakens AI Safety Commitments Amid Pentagon Ultimatum Over Military Use
Anthropic dropped its core AI safety pledge as the Pentagon set a Feb 27 deadline for unrestricted Claude access. What this means for the industry.
Popular topics
Topics with the most content and project reviews