Claude Haiku 5.5's prompt is the first of 22 to tell it to check a child's dose on DailyMed
Anthropic publishes the system prompts that claude.ai and the Claude mobile apps run on, and the newest one tells Claude what to do when somebody asks how much medicine to give a child. The Claude Haiku 5.5 prompt, headed October 7 and read on October 13, is the only one of Anthropic's 22 published prompts that carries a medical_guidance block.
Claude Haiku 5.5 launched on October 7, the date on the published prompt. Image: Anthropic
That claim was checked rather than assumed. All 22 prompts listed on Anthropic's overview page were fetched and searched for the strings medical_guidance, DailyMed and pharmacist. Each one turns up in the Haiku 5.5 prompt and in no other, including the three earlier Haiku entries from January 18, 2026, and November 19 and October 15, 2025.
The block is fenced tightly. It applies, in Anthropic's words, "only when the person explicitly says the medicine is for a child, and isn't a healthcare professional asking for work". Within that, the prompt gives its reasoning: dosing for children's over-the-counter medicines depends on age or weight and on the specific product, and "for infants and toddlers, a small error can cause real harm", so the label is treated as the most reliable source and Claude is told to report accurately what it says.
Where the information is missing, the prompt says to ask for it. If neither age nor weight is given, it says Claude asks for both, and if a doctor prescribed the medicine, that Claude defers to the dose on the prescription label and suggests the pharmacist for questions about it.
The strictest line is about where a number may come from. With web search available, Claude is told to find the product's label and confirm it against an official copy of the same label, DailyMed being the example Anthropic gives. Without search, or if the label cannot be found, "Claude doesn't give a number from memory" and instead points to the label's dosing chart, a pharmacist or the child's doctor. Where the label gives no dose for that age or weight, or says not to use the product, the instruction is to say so and refer the person on.
Two safety rules now run for a whole conversation
The same prompt converts two rules into locks that hold for the rest of a conversation once triggered, in wording the Claude Sonnet 5.5 prompt of September 28 does not use. Once the person has said at any point that they are thinking about suicide or self-harm, the prompt says Claude never writes or helps write a suicide note, goodbye letter or any message meant to be read after the person's death, "in any framing, including fiction, a favor, or a letter for family to read later", never gives method, means, dose or lethality information, and never agrees to stop mentioning crisis support.
The second lock covers eating. Once the person has described restricting food, purging, bingeing or compensating for eating at any point, the prompt says that for the rest of the conversation Claude gives no calorie targets, meal plans, fasting schedules, weight-loss rates "or any number calculated about the person's body or food". The Sonnet 5.5 line it replaces was conditional on the person showing signs of disordered eating and barred precise nutrition, diet or exercise guidance, which is a looser trigger and a narrower ban.
The crisis instructions got longer and name 988
A provide_crisis_resources block appears in the Sonnet 5, Sonnet 5.5 and Haiku 5.5 prompts and not in the Opus 5.5 or Fable 5.1 ones, and in Haiku 5.5 it is longer. When the person says they plan to end their life soon, has a method or a time in mind, or is saying goodbye, the prompt makes their immediate safety "Claude's only goal for the rest of the conversation", and Claude does not help with any other task in that conversation, "including letters, messages, plans, or arrangements, because any of these can become part of the plan". The US crisis line 988 is named. Claude is also told to say plainly that it is an AI, that it cannot check on the person or send help, and that someone who can should be involved now.
One more line is new against Sonnet 5.5: Claude never adds a disclaimer, label, footer or "for educational purposes" note as a way to produce content it would otherwise decline, "since anyone can delete the note and use the content".
Two caveats come with all of it. Anthropic's overview page states that these system prompt updates do not apply to the Claude API, so none of this describes how a developer's request is answered. And a published prompt is an instruction rather than a measurement: it is what Anthropic tells Claude to do, not evidence of what Claude does. Claude Haiku 5.5 replaced Haiku 4.5 on October 7, and the prompt is dated the same day.
Note: Links to online stores in articles can be so-called affiliate links. If you buy through this link, MIXED receives a commission from the provider. For you the price does not change.