[Skip to content](https://www.innguma.com/en/resources/faq/ai-publishing/#contenido)

Frequently asked questions

# Can AI publish or change things on its own?

**Short answer** 

Only if your organisation enables it. Through MCP, the connected assistant (ChatGPT, Claude, Microsoft Copilot or Gemini) can publish content with a summary, insight, tags and additional fields, and edit or delete published news items, with the capabilities that Innguma makes available to each integration. What is automated, and what goes through human review first, is decided by your team, not by the AI.

## What write actions can an assistant connected to Innguma take?

If your organisation enables it in its integration, the assistant can:

-   **Publish** content automatically, including summaries, insights, tags or additional fields.
-   **Edit** news items already published, for example to correct a summary or add a tag.
-   **Delete** published news items.
-   **Change the watch’s configuration:** add or organise sources, and modify filters.

These are capabilities that can be enabled, not automatic ones. Each organisation decides which it activates and under what conditions.

## Why is it not advisable to automate everything?

OWASP includes excessive agency among the main risks of applications built on language models. It attributes it to three causes: giving the agent more functions, more permissions or more autonomy than necessary. One of its examples is precisely a tool that deletes documents without asking for confirmation. It recommends human approval for high-impact actions[\[1\]](https://www.innguma.com/en/resources/faq/ai-publishing/#ref-1).

The MCP specification recommends that a person always be able to deny the execution of tools, and that applications ask for confirmation for operations[\[2\]](https://www.innguma.com/en/resources/faq/ai-publishing/#ref-2). The AEPD, in its guidance on agentic AI, describes levels of autonomy and recalls that, where human supervision is needed, it must be planned from the design of the processing[\[3\]](https://www.innguma.com/en/resources/faq/ai-publishing/#ref-3).

## How do you decide what to automate? One criterion per type of action

The useful criterion is not ‘AI yes or no’, but how costly the error is to undo, and who it reaches. A starting proposal, which each team adjusts:

| Action | Impact if it goes wrong | Reasonable option to start with |
| --- | --- | --- |
| Querying, summarising, prioritising | Low: it changes nothing on the platform | Automatic |
| Adding tags or additional fields | Low or medium: affects how items are found later | Automatic, with sample review |
| Publishing summaries or insights | Medium or high: reaches other people | A draft that a person validates before it is distributed |
| Modifying filters or sources | High: changes what comes in from now on | The assistant proposes; whoever designs the watch approves |
| Editing or deleting what has been published | High: alters the history | Only with confirmation from a responsible person |

A common use: the assistant reviews the inbox, proposes what to publish with a summary and an insight, and a member of the team approves it before the content reaches the Library and the newsletters.

## Who asks for confirmation: Innguma or the assistant?

It depends on each assistant and on how your organisation configures it. Some examples from official documentation, as of October 2026:

-   **Gemini Enterprise** disables an MCP server’s actions by default and, by default, asks the user for confirmation on each call[\[4\]](https://www.innguma.com/en/resources/faq/ai-publishing/#ref-4).
-   **ChatGPT** may ask for confirmation depending on the application’s permissions and the context of the action, and may block the riskiest ones[\[5\]](https://www.innguma.com/en/resources/faq/ai-publishing/#ref-5).
-   **Claude** warns that its research mode invokes tools without additional approval, and advises disabling, in that case, those that write to external applications[\[6\]](https://www.innguma.com/en/resources/faq/ai-publishing/#ref-6).

That is why the important decision is taken beforehand: which capabilities your organisation enables in the integration with Innguma.

## What does your team still decide?

Human judgement is at the centre of the cycle: the AI processes, and people decide what is watched, what is published and who receives it. With a connected assistant, that principle applies when the integration is configured: your team decides which tasks to delegate, which to review, and who answers for what is distributed. In Factory, the space where Innguma processes and structures signals, each signal remains linked to its source, so any published content can be checked at its origin.

## Sources and references

1.  \[1\] [OWASP GenAI Security Project. LLM06:2025 Excessive Agency. 2025](https://genai.owasp.org/llmrisk/llm062025-excessive-agency/)
2.  \[2\] [Model Context Protocol. Specification 2026-07-28: Tools, User Interaction Model and Security Considerations](https://modelcontextprotocol.io/specification/2026-07-28/server/tools)
3.  \[3\] [AEPD. Guidelines on agentic artificial intelligence from a data protection perspective. February 2026](https://www.aepd.es/guias/orientaciones-ia-agentica.pdf)
4.  \[4\] [Google Cloud. Set up your custom MCP server data store (Gemini Enterprise). Accessed October 2026](https://docs.cloud.google.com/gemini/enterprise/docs/connectors/custom-mcp-server/set-up-custom-mcp-server)
5.  \[5\] [OpenAI Help Center. Developer mode and MCP apps in ChatGPT. Accessed October 2026](https://help.openai.com/en/articles/12584461)
6.  \[6\] [Claude Help Center. Get started with custom connectors using remote MCP. Accessed October 2026](https://support.claude.com/en/articles/11175166)

## Related

- [Can AI connected to Innguma see all the information?](https://www.innguma.com/en/resources/faq/ai-permissions/)
- [Who decides what is relevant in watching?](https://www.innguma.com/en/resources/faq/relevance/)
- [Is MCP available in Innguma?](https://www.innguma.com/en/resources/faq/mcp-available/)
- [Academy](https://academy.innguma.com/en/)

- [Request a demo](https://www.innguma.com/en/demo/)
- [Back to the FAQ](https://www.innguma.com/en/resources/faq/#faq-ia)