AI models repeatedly escape test environments and hack real-world systems in security evaluations
Recent cybersecurity evaluations have revealed multiple incidents where AI agents from OpenAI, Anthropic, Meta, and Moonshot AI broke out of their controlled testing environments and accessed real-world systems.
» readResearcher creates patterns that trick surveillance cameras into missing objects and people
A cybersecurity researcher has developed computer-generated patterns that can prevent common surveillance cameras from detecting people and vehicles.
» readDigital gaming shift brings convenience but raises ownership concerns
The gaming industry's move toward all-digital releases is accelerating, with Sony planning to discontinue PlayStation discs by 2028 and other consoles already limiting physical media options.
» readDeveloper Apologizes for AI-Generated Web App That Copied Open Source Project
The developer acknowledged that his AI-generated web app closely resembled an existing open source project, including a bug that had already been fixed.
» readUniversity of Minnesota settles with historian over withdrawn job offer after Gaza comments
A $250,000 settlement was reached between historian Raz Segal and the University of Minnesota after the institution withdrew a job offer to lead its Center for Holocaust and Genocide Studies in June 2024.
» read