Security & Data-Privacy Compliance

AEGIS (“Adaptive Engine for Guided Investment Study”) helps students understand markets. A student asks a question about a stock, crypto asset, or options contract (for example, “What’s the read on TSLA?”) and AEGIS returns an educational, institutional-style briefing. All trading on the platform is simulated; AEGIS provides analysis for learning, not financial advice.

When a student submits a query, our backend assembles a payload for the language model that contains only the following:

• The market question the student typed (e.g. a ticker symbol or a sentence about a market).
• Public market data we fetch from financial data providers: prices, trading volume, technical indicators (RSI, MACD), market capitalization, options data (Greeks, open interest), macro snapshots (S&P 500, US Dollar Index, 10-year Treasury yield), and public news headlines for the asset.
• A fixed system instruction that defines AEGIS’s analytical persona and output format.
• For follow-up questions in the same chat, the prior market questions and AEGIS answers in that conversation, used purely as context.

That is the complete list. The payload is constructed deterministically by our code from market-data feeds — there is no field in it for a student’s name, email, ID, grade, school, or any other personal identifier, and none is added.

Students must be signed in to use AEGIS, and we store each conversation so a student can revisit it. We do this using an internal account identifier (a random user ID) that lives only in our own database. That identifier is used to scope and retrieve history on our servers; it is not included in the request to the external AI model. The link between a conversation and a real student exists only inside our SOC 2-certified database, never at the AI provider.

AEGIS uses xAI’s Grok model through xAI’s business API (api.x.ai), governed by xAI’s enterprise terms — not the consumer Grok product. Under those terms, business API inputs and outputs are not used to train xAI’s foundation models in the way consumer chats may be, and xAI offers a Zero Data Retention (ZDR) option for enterprise accounts under which prompts and completions are processed in real time and never persisted. Because our payload already excludes all student PII, even the data that does transit to the model contains no education record.

Consistent with NYCPS AI guidance, AEGIS:

• Is clearly disclosed to schools as an AI feature, with this documented data flow.
• Produces explanations a student and teacher can read and understand — it states a verdict and the reasoning, not an opaque score alone.
• Does not make consequential decisions about students (no grading, discipline, or eligibility decisions). It is an educational explainer.
• Is held to the same privacy and security standards as the rest of the platform; if we add or change AI features, we will notify NYCPS as required.
• Carries an explicit disclaimer that its output is algorithmic market analysis for educational use, not financial advice.

For a single AEGIS request, data moves as follows:

• 1. Student types a market question in the Learning Hub and is authenticated.
• 2. Our backend fetches public market data for the asset from financial-data providers.
• 3. A deterministic engine computes technical indicators from that market data.
• 4. We send the market question + market data to xAI’s Grok API to write the narrative analysis. No PII is included.
• 5. The streamed answer is returned to the student.
• 6. The question and answer are saved to our database, scoped by internal user ID, so the student can revisit the conversation. The student can delete a conversation at any time.