← Bridge Trading

The Analyst (AI) Disclosure

Document hash: b30c02bc40ed · Bridge Trading legal corpus
⚠ DRAFT — v0, not yet effective. This document is a working draft prepared by the founder. It has not yet been reviewed by counsel and is not effective until lawyer-approved and posted with a confirmed Effective Date. Customer-facing reliance on this language is at your own risk pending finalization.

Bridge Trading — The Analyst (AI) Disclosure

Notice to Reader (this version)

This is a v0 working draft prepared by the founder. It consolidates Bridge Trading's full position on artificial-intelligence-generated content, large-language-model use, and training-data policy in one document so customers, regulators, and reviewing counsel can find everything about AI in one place.

This disclosure is incorporated into the Terms of Service (§9), Privacy Policy (§9), and Acceptable Use Policy (§5(c)) by reference.

1. What "The Analyst" Is

"The Analyst" is the customer-facing name for the artificial-intelligence-powered research assistant feature of the Bridge Trading Service. It is implemented using third-party large language models, currently provided by Anthropic (the Claude family of models), accessed via the Anthropic commercial API.

The Analyst surfaces in several places in the Service:

Everywhere The Analyst's output appears in the Service, the source of the content is labeled "AI-generated" with a visible icon or banner.

2. The Analyst is NOT a Human

This is the most important sentence in this document, so it gets its own section:

The Analyst is a software program that generates text based on statistical patterns learned during its training. It is not a licensed financial professional. It is not your fiduciary. It does not know your specific financial situation. It is wrong sometimes — including in ways that sound completely confident.

Specifically, The Analyst:

(a) Is not a registered investment adviser, broker-dealer, commodity trading advisor, certified financial planner, certified public accountant, or any other regulated financial professional. (b) Is not reviewed by a human licensed financial professional before output is delivered to you. (c) Does not know your account balance, your risk tolerance, your tax situation, your other holdings, your time horizon, your dependents, or any other personal circumstance unless you have explicitly told it within a given conversation. (d) Has a training-data cutoff beyond which it does not know about recent events. The current model knows information up to its training cutoff date; for events after that, The Analyst will either say it doesn't know or, in some cases, generate plausible-sounding but incorrect content (called "hallucination").

3. Known Limitations You Should Treat as Defaults

When you read or use Analyst output, assume the following are true unless explicitly indicated otherwise:

3.1 Hallucinations

Large language models occasionally generate confident-sounding statements that are factually wrong. Examples:

This is a known characteristic of the technology, not a bug. Do not act on a specific factual claim from The Analyst without independently verifying it.

3.2 Stale information

The Analyst's training data has a cutoff date. For any event after that date, The Analyst may either acknowledge the cutoff, attempt to extrapolate (poorly), or simply make something up. Bridge Trading attempts to inject real-time data (current quotes, today's GEX levels, this morning's news) into Analyst prompts where possible, but the underlying model's general knowledge is still date-limited.

3.3 Bias toward fluency over accuracy

Large language models optimize for output that sounds good, not for output that is accurate. A confident-sounding paragraph from The Analyst is not, by itself, evidence that the underlying claim is correct.

3.4 Math errors

Large language models are known to make arithmetic errors, especially with multi-step calculations involving options Greeks, position sizing, P&L projections, or risk/reward ratios. Bridge Trading does its calculations in code, not in the LLM. If The Analyst quotes a number, treat it as illustrative; verify against the in-Service calculator.

3.5 Cross-conversation forgetting

The Analyst does not have persistent memory of your prior conversations unless that memory is explicitly stored and re-injected (which Bridge Trading does for certain types of context, such as your watchlist). Each new conversation starts mostly fresh.

3.6 No real-time market data unless explicitly injected

If you ask The Analyst "what is SPY trading at right now?" — the answer is only reliable if Bridge Trading has explicitly injected a fresh quote into the prompt context for that conversation. Bridge Trading does inject quotes for tickers in the dashboard view, but not for all possible queries. When unsure, check the actual Service ticker bar, not The Analyst.

4. What The Analyst is NOT for

Do not use The Analyst as:

5. What The Analyst IS Good At

Used appropriately, The Analyst is genuinely useful for:

The pattern: use The Analyst for generation, framing, and explanation. Verify against actual data and verified sources for facts and numbers.

6. Bridge Trading's Commitment: We Never Train On Your Data

Bridge Trading commits to never training any artificial-intelligence model on customer-specific content. This commitment has the following specific components:

6.1 What we never use as training data

None of this content is used as input to training, fine-tuning, or reinforcement-learning processes for any foundation model, customized model, or internal model operated by Bridge Trading or by any subprocessor on our behalf.

6.2 What we DO use, with anonymization

Bridge Trading uses aggregated, anonymized usage statistics to operate, maintain, and improve the Service. Examples of acceptable anonymized use:

These types of aggregated statistics are used internally for product improvement and capacity planning, and may be shared in marketing materials in clearly anonymized form.

6.3 What our third-party AI provider does with your prompts

When you submit a prompt to The Analyst, Bridge Trading transmits the prompt to Anthropic's commercial API for processing. Our contract with Anthropic prohibits Anthropic from using customer API content as training data for their public foundation models. This commitment is reflected in Anthropic's commercial-API terms of service.

Anthropic may retain prompts and responses for a short operational window (typically 30 days) for abuse detection and incident investigation. After that window, the content is deleted per Anthropic's data retention policy. We do not control Anthropic's internal retention; we have selected Anthropic specifically because of the strength of their commercial-API privacy posture compared to peer providers.

6.4 If we ever want to change this commitment

If Bridge Trading ever wishes to change the no-training commitment in Section 6.1 — for example, to enable training of an internal Bridge-specific model on customer prompts — we will:

(a) Notify every then-active customer by email at least thirty (30) days before the change takes effect; (b) Allow you to opt out of the change without affecting your access to the Service; (c) Permit customers who do not wish to consent to the change to cancel their Subscription within the notice period and receive a pro-rated refund of the unused portion of the then-current month; (d) Update this document and post the prior version in a public changelog.

We have no current intention to change this commitment. The notification process above exists so that, if commercial realities ever shift, customers have advance warning and an opt-out path, rather than discovering the change after the fact.

7. How We Decide Which Model to Use

Bridge Trading currently uses Anthropic's Claude family of models. The specific model and version may change as Anthropic releases updates. Considerations we weight when selecting a model:

(a) Privacy and training-data policy — we select models from providers who contractually commit not to train on commercial-API customer content. (b) Reliability of refusals on out-of-scope requests — we want a model that consistently declines to play "your investment adviser" when prompted. (c) Safety performance on financial topics — we want a model that does not casually offer specific buy / sell recommendations dressed up as "general advice." (d) Hallucination rate on trading-relevant topics — measured via our internal eval set. (e) Latency and cost — within the acceptable-quality envelope.

If we materially change AI providers or models (e.g., adding OpenAI, swapping to a different family), we will update this document and notify customers.

8. Special Categories of Output

8.1 Performance numbers The Analyst quotes

If The Analyst quotes a "win rate," "average return," "Sharpe ratio," or any other performance number for a Signal Provider, a strategy, or a backtest:

8.2 Trade ideas The Analyst suggests

If The Analyst suggests a specific trade — "consider a SPY 450/460 call spread for July expiry" — that suggestion is:

The Service is designed so that no order is ever submitted automatically based on Analyst output. Order submission requires either a Signal Provider's dispatched signal (which the Customer has separately chosen to follow) or a manual Customer action.

8.3 Risk-management advice

If The Analyst offers risk-management guidance — "use a 1-2% per-trade position size" — treat it as:

If you need risk-management guidance calibrated to your actual financial situation, consult a licensed adviser.

If you ask The Analyst about tax treatment (e.g., "is this a wash sale?"), accounting (e.g., "how do I report this on Schedule D?"), or legal questions (e.g., "is this OK under my prop firm's rules?"):

9. Logging and Audit

Bridge Trading logs the following for each Analyst interaction, for the purposes described in the Privacy Policy:

This logging:

You can view your own Analyst conversation history in the Service dashboard. You can request a copy of all logged conversations under the data-portability rights described in the Privacy Policy.

10. Reporting Bad Analyst Output

If The Analyst gives you output that is:

…please report it to [email protected] with a description of what happened and (if possible) the conversation thread ID.

We treat these reports as important quality signals. We do not retaliate against good-faith reporters.

11. What Changes Trigger an Update to This Disclosure

We will update this disclosure (and notify customers) when any of the following materially changes:

The "Last Updated" date at the top of this document indicates when this disclosure was last revised. Material changes are also communicated by email to your account email address with at least thirty (30) days' notice.

12. Contact

Questions about The Analyst, the AI training policy, or anything in this disclosure: [email protected]

Reports of bad Analyst output: [email protected] with "AI Output" in the subject line

[Bridge Trading LLC] Attn: AI / Analyst [REGISTERED_AGENT_ADDRESS — TBD]

This document is a v0 working draft prepared by the founder. It has not yet been reviewed by counsel and is not effective until lawyer-approved and posted with a confirmed Effective Date.