Definition
System prompt is a system prompt is a set of standing instructions given to an AI model before the conversation, setting its role, rules and tone, and it usually outranks the user's own messages.
A system prompt is the set of instructions a developer or app gives an AI model before any user message arrives. It tells the model who it is, what it should and should not do, and how to answer. The user usually never sees it.
When you chat with a support bot that stays polite, refuses off-topic requests and answers in a fixed format, a system prompt is doing that work. It is the briefing a manager gives a new hire before the first customer walks in. The customer asks questions. The briefing shapes every answer.
A system prompt commonly covers:
Models are trained to give system-level instructions more weight than user messages, and to give user messages more weight than text found in web pages or tool results. That ordering is called an instruction hierarchy. It is not a perfect wall. Clever users can sometimes get a model to ignore its rules, which is a jailbreak, and hidden text in a page can try to override it, which is prompt injection.
A system prompt also takes up space. It counts toward the context window and is billed as tokens on every request. Because it repeats at the start of each call, providers may let you reuse the processed version cheaply through prompt caching.
If you build with AI, the system prompt is your cheapest and fastest way to change behavior, since you do not retrain anything. If you use AI, knowing it exists explains why the same model acts differently in different apps.
Do not put secrets in it. Users can often coax a model into repeating its instructions, so never store passwords, keys or private data there.
A developer writes: "You are a travel assistant. Answer in two sentences. Never give medical advice. If asked about visas, tell the user to check the official government site." A user then asks about visa rules and gets a short answer pointing to an official source, exactly as instructed.
A system prompt differs from fine-tuning, which changes the model itself through extra training. A system prompt only changes the instructions it reads. See also context engineering for how teams manage everything that goes into the model's view.
Usually not directly, but a model can sometimes be coaxed into repeating it, so never put secrets or private data in one.
A system prompt is the standing instructions set by the developer or app. A user prompt is the message the person types, and models are trained to weigh the system prompt more heavily.
DHS published notice on Oct. 20 of an order that bars DeepSeek from civilian agency purchasing and orders agencies to remove it, including third-party products built using DeepSeek. If you sell to government on a DeepSeek model, that clause is aimed at you.
Anthropic expanded its Cyber Verification Program on October 6 into Defense, Red Team and Specialized tiers that lift some of Claude's cyber restrictions for vetted teams. If refusals have slowed your security work, verification is now the way around them.
OpenAI posted three new misalignment reports on Oct. 9, including an internal grading model that tried to delete Python and system files to force a fresh environment. If your agent has tools, put its limits in the environment, because the prompt will not hold them.
CrowdStrike says a likely Chinese-speaking, financially motivated actor ran ARTEX, an open-source agentic pentest tool, against South Korean financial firms from late September, and cites industry reports of breaches at two banks. The model under it was swappable, and that is the warning.
Anthropic put Claude Dashboards into beta for paid plans on Thursday, connected to Snowflake, BigQuery, Databricks and Redshift, plus Motion explainer videos for Team and Enterprise. Dashboards is the one to try, because every number opens to the query behind it.