AI Transparency Statement
Last updated: July 2026
This page discloses the AI model powering StrategicAi, the data sources feeding live briefs, what we can and cannot know, and our commitments on query privacy. It is written for analysts, security officers, and procurement teams evaluating the platform.
1. Model & Provider
Model: Claude by Anthropic, routed by plan — Claude Haiku 4.5 (claude-haiku-4-5) on free, trial, and Operator briefs; Claude Sonnet 5 (claude-sonnet-5) on Analyst and the public v1 API; Claude Opus 4.8 (claude-opus-4-8) on Desk and BYOK. The badge on each brief names the model that produced it
Provider: Anthropic, PBC — anthropic.com
Extended thinking: Enabled for Predict / Forecast mode, with a 6,000-token reasoning budget. This allows the model to work through multi-step geopolitical scenarios before producing a probability estimate.
Knowledge cutoff: August 2025. Model knowledge is supplemented by live RSS news feeds and FX market data fetched at brief-generation time.
API governance: Usage is governed by Anthropic’s usage policies and privacy policy.
2. Data Sources
Each live brief draws on the following sources. Where a source is used, the relevant item carries an inline citation tag.
| Source | Type | Refresh | Coverage |
|---|---|---|---|
| Live RSS feeds | News headlines | Every 25h | BBC News, CNBC, NYT Business, MarketWatch |
| Frankfurter (ECB) | FX rates | Hourly | EUR/USD, GBP/USD, USD/JPY |
| ACLED | Conflict events | 6h cache (updates weekly) | 50+ countries, village-level geocoding |
| Claude (Anthropic) | Background knowledge | Haiku 4.5 / Sonnet 5 / Opus 4.8 by plan | Global geopolitics, economics, markets |
| User context | Analyst profile | Session | Provided by user in Desk — not verified |
3. What We Know, What We Don’t
Strong signal
- • Macro geopolitical trends and named actors
- • Historical context and precedent
- • Event-class probabilities from historical base rates
- • Structural economic and market dynamics
Weak / not available
- • Day-of breaking news (RSS up to 25h delayed)
- • Private company financials
- • Classified or non-public information
- • Bloomberg Terminal, real-time equity prices
- • SEC EDGAR filings, social media signals
Every factual claim in live briefs is tagged with its source. Claims tagged [Analyst Assessment] are based on model knowledge, not live data.
4. Query Privacy
These are hard commitments, not aspirational policy:
Your queries are NOT used to train any AI model. This is guaranteed by Anthropic’s Enterprise API terms — we operate under a data-processing agreement that prohibits training use.
Query content is not shared with third parties except for the sole purpose of generating your response (i.e., the API call to Anthropic). No analytics vendors, data brokers, or advertising networks receive your queries.
Query logs are retained for 30 days for security and rate-limiting purposes, then permanently deleted.
Signed-in users can request deletion of their query history at any time — email privacy@strategicglobal.app.
5. Output Limitations
Institutional buyers make better decisions with honest disclosures. Here is what you should know:
- • Probabilistic, not advisory. AI outputs are probabilistic estimates. They are not investment advice, legal counsel, or operational intelligence.
- • Forecasts can be wrong. Probability forecasts are calibrated against historical accuracy, but individual forecasts carry real uncertainty. A 70% probability is wrong 30% of the time by definition.
- • Hallucination risk is real. The model can generate plausible-sounding claims that are incorrect. Inline
[Source, Date]citations indicate grounded claims;[Analyst Assessment]indicates model inference. Verify grounded claims against primary sources before acting. - • Not a replacement for primary research. Do not rely on StrategicAi outputs as the sole basis for material financial or security decisions.
- • No privileged access. StrategicAi does not have access to non-public information, insider knowledge, or classified government intelligence.
6. Calibration & Track Record
Every forecast we publish is logged at the time of generation. Brier scores and hit rates are published as outcomes resolve. This is the accountability standard we hold ourselves to — not cherry-picked wins.
7. Data Processing Addendum (DPA)
For enterprise and Desk-tier customers requiring a DPA for procurement compliance, email legal@strategicglobal.app. We will provide a DPA template covering:
- • Data controller / processor roles
- • Sub-processors (Anthropic, Vercel, Upstash)
- • Retention periods and deletion procedures
- • Data subject rights and breach notification obligations
8. Support the Free Tier
The free briefs you run here — like the one that probably brought you to this page — are real Claude calls, paid for out of pocket so that anyone can access institutional-grade intelligence, not just those who can expense it. If StrategicAi has been useful to you, you can help keep the free tier free:
Donate any amount via Cash App: $SGlobalApp
Every dollar goes back into the product — model costs for free users and building what’s next. Nothing is pocketed.
And if you’ve already given: thank you. You’re not just covering API bills — you’re backing a dream of putting this kind of intelligence in many more hands.
Donations are voluntary and buy nothing — no access, no features, no priority. For product access, see the plans on the home page.