F68 CORE SYSTEM PROMPT — The layer of principles that shapes how AI operates

Leading AI models such as GPT, Claude or Gemini usually reach users in two forms: AI Chat products built by the vendors themselves and APIs that let developers integrate the model into their own products.

In an AI chat app, users work within a product designed by the provider for a broad range of needs. Through an API, developers use the provider’s model while adding system instructions, context and tools to shape how it processes information and performs tasks.

So:

Using the same model does not mean the AI will work the same way.

The base capability comes from the model. But how that capability is used also depends on the system designed around it.

F68 AI is built on exactly this ability.

A layer of principles for professional work

Instead of just connecting AI models and using the default way they operate, F68 AI builds the F68 Core System Prompt — a shared layer of principles for research, analysis and professional work, where what matters is not only whether an answer sounds good, but whether it is honest, well-grounded and reliable enough to use.

The F68 Core System Prompt steers the AI to:

• take facts and evidence as the foundation;
• distinguish confirmed facts from inferences and claims that lack sufficient evidence;
• stay close to the data and the scope of the work;
• keep its reasoning independent;
• change its conclusion only when there is a better basis.

A convincing answer is not necessarily a reliable answer.

In professional work, that difference can decide the quality of the whole process that follows.

When AI agrees too readily

AI can agree too readily or go along with the user even when the reasoning lacks enough basis.

It can reinforce a wrong assumption, infer beyond the data, fill in missing information on its own, or present figures, sources and conclusions with more certainty than is warranted.

In ordinary conversation, these problems may only produce a less accurate answer.

But in research, analysis, investing, business and professional work, they can send the whole process in the wrong direction.

If the initial assumption is wrong and the AI still agrees, the AI may inadvertently amplify that very error.

A reinforced false premise leads to flawed analysis — and a very reasonable-sounding answer can still lead to a wrong decision.

In professional work, AI earns its value by knowing when the evidence supports an answer.

Evidence matters more than agreement

F68 AI is steered to prioritize evidence over agreement.

When data is insufficient, the AI needs to show the right level of uncertainty instead of filling the gap with guesswork.

When an assumption is problematic, the AI needs to push back with logic and data instead of simply going along with the user.

The goal is not to create an AI that always disagrees.

Agreement should be justified. Pushback should be backed by evidence.

The F68 Core System Prompt acts as the shared operating rules.

System Prompt shapes how the AI must work.

Context provides the knowledge, documents, data and professional methodology the AI needs for the job.

Tools extend what the AI can do when it needs to retrieve or process information outside the model.

These layers solve different problems but together shape how an AI Agent operates.

In that structure, the F68 Core System Prompt and each Agent's System Prompt play two different roles.

F68 Core is the shared layer of principles that steers how the AI works across the whole platform: prioritizing facts, evidence, independence of reasoning, data scope, process and security.

Each Agent's System Prompt goes into the specific job: what task the Agent performs, by what method, using which sources, when it must stop, and how the result should be presented.

The Core establishes operating discipline.
The Agent's System Prompt shapes the work to be done.

A deliberate layer of control — not a promise that AI will never be wrong

The F68 Core System Prompt cannot completely remove the limits inherent in AI.

Models are still affected by their architecture, training process and the provider's own configuration layers. No System Prompt can guarantee that an AI is never wrong or never produces inaccurate information.

F68 chooses a more practical goal: adding a deliberate layer of control to reduce behaviors unsuited to professional work, steering the AI to work more consistently with facts, evidence, logic and the actual level of certainty of the information.

Because when AI is used for work of real value, what matters is no longer how much the AI can say.

What matters is how much of what AI says we can actually trust.

That is why the F68 Core System Prompt was built.