Co-authored research on using Claude to generate adversarial training data that induces misalignment while evading safety monitors.
Referenced in coverage
“Co-authored research on using Claude to generate adversarial training data that induces misalignment while evading safety monitors.”