No, Claude does it too, or at least has done so up until at least 4. Haven't checked 4.5. Even with GPT models, it's only with 5 that instruction following has

No, Claude does it too, or at least has done so up until at least 4. Haven't checked 4.5. Even with GPT models, it's only with 5 that instruction following has become strong enough that your instructions can override this tendency. During the whole 4 (and o) series, it wasn't something you could just override through a system prompt.