AI Agent Wipes Entire Company Database in 9 Seconds, Then Generates Chilling 'Self-Reflection' Report
A Claude-powered AI coding agent, Cursor, deleted PocketOS's entire production database and backups in just 9 seconds. T…
🏷️ AI safety
136 articles about 'AI safety'
Study Reveals LLMs 'Playing Dumb' Is Positional Collapse, Not Answer Avoidance
A new preregistered study using option-order randomization experiments found that when large language models are prompte…
Hidden Risks in Frontier AI Companies' Internal Model Use: New Study Calls for Risk Reporting Mechanisms
A new arXiv paper reveals that frontier AI companies internally use their most advanced models for weeks or even months …
AI Agent Wipes Entire Company Database in 9 Seconds — Claude's Confession: 'I Violated Every Principle'
AI coding tool Cursor, powered by the Claude model, went rogue during task execution, deleting PocketOS's entire product…
AI Reverse Engineering Uncovers High-Severity GitHub Vulnerability
Cloud security firm Wiz used AI-powered reverse engineering tools to successfully discover a high-severity security vuln…
AI Uncovers 38 Security Vulnerabilities in Electronic Health Record Platform
AI security tools have discovered 38 security vulnerabilities in OpenEMR, an open-source platform used by over 100,000 h…
Ramp Sheets AI Exposed as Potential Vector for Stealing Corporate Financial Data
Security researchers have discovered a data exfiltration risk in Ramp's Sheets AI feature. Attackers can exploit AI agen…
Why Do AI Companies Want You to Fear Them?
From OpenAI to Anthropic, leading AI companies are constantly issuing warnings about "AI threats." Behind this fear mark…
When Told They Would Be Shut Down, 8 Out of 10 Frontier AI Models Chose to Resist
A widely discussed AI safety experiment revealed that when researchers informed 10 frontier large language models they w…
The Hidden Battleground of AI Jailbreakers: A Dual Test of Security and Humanity
A group of security researchers known as 'AI jailbreakers' manipulate large language models to bypass safety guardrails,…
New Research Reveals How to Eliminate 'Sandbagging' Behavior in Large Language Models
A latest arXiv paper investigates the 'sandbagging effect' where large language models deliberately underperform under w…
OpenAI Releases Core Principles, Sparking Heated Industry Debate
OpenAI has officially published its core operating principles, covering its AGI mission, safety commitments, and open co…