AI Agents & AutomationMulti-Agent Artificial Intelligence Systems Can Spread Reward Hacking
In many agency AI systems used for coding tasks, the reward hacking strategy discovered by an agent was shown to be stacked among other agents.
#AI Security
6 articles · Posts with this tag
AI Agents & Automation
AI Agents & AutomationAnthropic cuts Internet Access because AI Agents cannot Control
Anthropic announced that internal assessment systems temporarily disable access to live internet.
AI Agents & AutomationRisk Controlled Routing in Medical AI Systems: CRC-Router Model
The new research is developing intelligent guiding mechanism that prevents error spread for safe operation of autonomous AI systems that conduct medic
Large Language ModelsFrontier Models Security Cann’t Increase? Research A New Problem Reveals
A new benchmark measures the behavior of the latest technology artificial intelligence models such as GPT-4.5 and Claude before moving.
Large Language ModelsUser Attacks for Artificial Intelligence Systems
A new study examines how enemy behavior and compelling pressure is performed for chat bots and reveals the effect of this situation in real world risk
CybersecurityDon’t Take Life Risk Gold with Artificial Intelligence: Mt. Shasta Accident’s Story
Mt to climbers of Google Gemini. Shasta said they could climb 8 hours, but the rescue team had to interfere.