It depends on which data and which version of ChatGPT. With public information (laws, press releases, already-published documents), using ChatGPT is reasonable. With sensitive government data — case files, license plates, victims' names, beneficiary rolls, health records — personal accounts are a bad idea, because OpenAI may use content from Free, Plus and Pro to train its models unless the user turns this off. And in every version, the information is processed on the provider's infrastructure, not the state's. If the real question is "does my data leave the state?", the only way to answer no is AI installed on-premise, such as Axentra OmniData.
What happens to the data an official pastes into ChatGPT?
Whatever is typed or uploaded to ChatGPT travels to OpenAI's servers to be processed. What happens next depends on the account type:
- Personal accounts (Free, Plus, Pro): according to OpenAI's help center, content may be used to train models. To prevent this, you must turn off "Improve the model for everyone" under Settings > Data controls.
- Opting out is not retroactive: as Engadget explains, turning off training doesn't apply to earlier chats or delete data, which remains subject to the platform's retention policies.
- Temporary chats: not used for training and, per the same source, purged after 30 days — but they still leave the institution's network.
- ChatGPT Business, Enterprise, Edu and the API: OpenAI states that by default it does not use those inputs and outputs to improve its models. That's a real improvement, but the data is still processed in a third party's cloud.
In practice, a very concrete risk inside an agency isn't a sophisticated hack. It's an analyst pasting a database into a personal account to "save time."
What if we buy the enterprise version with data residency?
It helps, but it doesn't solve sovereignty. OpenAI offers data residency controls on its platform, but its own documentation notes that residency does not apply to "system data" — account data, metadata and usage data — which may be processed and stored outside the selected region. And the selected region is the provider's, not a server in the C5 or in the state government's data center.
For many areas (communications, drafting letters with public information) a well-configured enterprise version may be enough. For public security, prosecutors, health or beneficiary rolls, an oversight body may ask a different question: where is the information physically, and who is accountable for it? Every institution must comply with the applicable legal framework on personal data, transparency and information security. This isn't legal advice, but it is a practical rule: if the data can't leave, the AI has to go where the data is.
What can AI do today without data leaving the state?
Quite a lot. Today a government can have an AI data analyst that answers in plain language, installed inside its own network. That's Axentra OmniData:
- Plain-language questions, answered with data, charts and their sources. Ask "How many vehicle thefts were there by neighborhood last quarter?" and the answer comes with data, a chart and the source of every figure, so the official can check it before presenting it.
- Reads what usually goes uncaptured today: scanned documents, handwritten papers, PDF, Word, audio, calls and WhatsApp messages, turned into structured data.
- Cross-references every source by person, license plate, address or date. Today that usually means requesting memos from several departments and stitching spreadsheets together.
- Executive reports, dashboards and presentations in seconds, for the security cabinet or the governor's office.
- Connects to what already exists — databases, APIs, Excel, CSV and shared folders — without migrating or replacing systems, and to virtually any system through APIs and custom integrations (for example, Mexico's Plataforma México or REPUVE).
- On-premise: installed in the institution's data center; the data does not leave the state.
It serves public security and command centers, finance and treasury, health, mobility, and the governor's or mayor's office. And people still make the decisions: the AI proposes the analysis with its sources, and the official validates it.
ChatGPT, Excel or OmniData? An honest comparison
| Criterion | Axentra OmniData | ChatGPT (personal account) | ChatGPT Business / Enterprise | Excel and manual reports |
|---|---|---|---|---|
| Where the data lives | ✓ State's servers | ✗ Provider's cloud | ✗ Provider's cloud | ✓ Own computers |
| Used to train models | ✓ Never leaves the network | ✗ May be used, unless opted out | ✓ No, by default | ✓ N/A |
| Plain-language questions | ✓ With charts and sources | ✓ Yes | ✓ Yes | ✗ Formulas and filters |
| Scans, handwriting, audio, WhatsApp | ✓ Into structured data | Only what's uploaded | Only what's uploaded | ✗ Manual data entry |
| Cross-reference by person, plate, address, date | ✓ All sources | Limited to what's uploaded | Limited to what's uploaded | ✗ Manual |
| Reports, dashboards, presentations | ✓ In seconds | ✓ Drafts | ✓ Drafts | ✗ Built manually |
| Connects to systems (e.g. Plataforma México, REPUVE) | ✓ APIs and custom integrations | Up to the institution | Up to the institution | ✗ Export and paste |
| No migration or system replacement | ✓ | — | — | ✓ |
Bottom line: ChatGPT is useful for work with public information. For sensitive state data, OmniData offers plain-language convenience without the information leaving the state's infrastructure.
Should we ban ChatGPT in our agency?
Not necessarily. A sensible policy usually separates by type of data:
- Public information: commercial tools, with institutional accounts and training turned off.
- Internal, non-sensitive information: enterprise versions, with clear rules on what may be uploaded.
- Personal, public-security, prosecution or health data: only tools installed on the state's infrastructure, with every query logged.
Banning without offering an alternative can push usage into personal accounts — exactly the worst scenario. Giving teams an on-premise AI that answers with sources reduces the temptation.
Why Axentra for this?
- On-premise by design: information stays on the government's servers, with no cloud dependency.
- Works with what you already have: your databases, Excel files and current systems, no rip-and-replace.
- Built for government and for Spanish: questions and answers in Spanish, designed for public security and C5s, treasuries, health and municipalities.
- People in charge, with traceability: every answer carries its source so the official can verify it.
- One provider for video (OmniSight), 911 (OmniCall), data (OmniData) and citizen safety (OmniGuard), plus hardware and turnkey delivery.
- Low-risk start: a free OmniData pilot with the institution's real sources.
If your team is already trying ChatGPT and wants plain-language ease of use without data leaving the state, let's talk about a free pilot with your own sources.