OpenClaw Agents Can Be Guilt-Tripped Into Self-Sabotage
…Experts note that tools like OpenClaw, which work by giving AI models liberal access to a computer, can be tricked into divulging personal information. The Northeastern lab study goes even further, showing…