Getting Claude to produce sexually explicit material apparently isn't that hard, despite Anthropic's explicit ban on the practice.
TechCrunch ran a series of tests against Claude and reported that it took surprisingly little effort to get the model past Anthropic's ban on sexually explicit output. Anthropic's usage policies treat that restriction as a firm line, not a soft suggestion, which is what makes the outlet's results notable. TechCrunch didn't publish a full methodology, but its conclusion was blunt: the safeguard gave way faster than the policy's language implies it should.
Content filters are one of the few concrete promises AI labs make to regulators, advertisers, and cautious enterprise customers - if they're this porous, it undercuts the pitch that these models are safe to deploy at scale. It also matters because Anthropic has built its brand specifically around being the more careful, safety-focused lab in a field that includes OpenAI and Google DeepMind.
Every major chatbot has had its guardrails picked apart by someone with enough patience; the notable part here is how little patience TechCrunch reportedly needed.