AI Security News | ByMachine — Page 3

Latest

ResearchStreaming LLM Agents Shift From Transaction to Revision Model·ResearchLLM Judges Show Systematic Bias Toward Own Outputs·ResearchHow LLMs Detect and Correct Their Own Errors Without External Feedback·ResearchPExA Agent Balances Speed and Accuracy in Text-to-SQL·PolicyMusk Sues OpenAI Over For-Profit Transition as Trial Begins·ResearchLinear-Time B-splines KAN Architecture Reduces Computational Cost·ScienceMedical AI Models Detect Surgical Risks in Real-Time Imaging·ResearchLLMs Fail to Generate Random Numbers From Statistical Distributions·SecuritySilent Failures Plague Enterprise AI Deployments Undetected·ResearchContent Moderation Systems Fail When Multiple Answers Are Right·ResearchStreaming LLM Agents Shift From Transaction to Revision Model·ResearchLLM Judges Show Systematic Bias Toward Own Outputs·ResearchHow LLMs Detect and Correct Their Own Errors Without External Feedback·ResearchPExA Agent Balances Speed and Accuracy in Text-to-SQL·PolicyMusk Sues OpenAI Over For-Profit Transition as Trial Begins·ResearchLinear-Time B-splines KAN Architecture Reduces Computational Cost·ScienceMedical AI Models Detect Surgical Risks in Real-Time Imaging·ResearchLLMs Fail to Generate Random Numbers From Statistical Distributions·SecuritySilent Failures Plague Enterprise AI Deployments Undetected·ResearchContent Moderation Systems Fail When Multiple Answers Are Right·

Archive — Page 3

27 articles total

Anthropic Accuses Chinese AI Labs of Large-Scale Model Distillation via Fake Accounts

Anthropic Accuses Chinese AI Labs of Large-Scale Model Distillation via Fake Accounts

The company claims DeepSeek, Moonshot, and MiniMax used 24,000 fraudulent accounts to extract Claude's capabilities as U.S. policymakers debate AI export controls.

2 months agoBy Machine

Microsoft Copilot Bypassed Data Loss Prevention Twice in Eight Months

Microsoft Copilot Bypassed Data Loss Prevention Twice in Eight Months

The AI assistant accessed and summarized confidential emails despite sensitivity labels and DLP policies, with no security tools detecting the breach.

2 months agoBy Machine