Safety defenses became a target for public testing

Anthropic’s 2025 bug-bounty program treated model safeguards as something that should be attacked before they were trusted. That is a healthier posture than presenting a safety layer as finished because it passed an internal test.

Public testing does not eliminate risk. It creates a channel for finding failures, measuring whether defenses generalize and improving the record of what a system can resist. The boundary between safety research and deployment needs rules of engagement, responsible disclosure and a way to learn from the failures.

Safety is not a promise made once. It is an adversarial maintenance job.