Den här texten återges på Svenska i nästa kommentar.
There is a generally understandable skepticism around the use of AI - not only from me, but from others as well.
A discussion in the vBulletin forum raised some important concerns about security, data protection and how forum content is handled when AI tools are involved. In our case, the current OpenAI integration is still an early prototype. We have not yet built the full privacy and consent layer around forum content, and that is exactly why this feedback is important.
The intention is not to silently send full forum discussions to an external AI provider, especially not when the forum are hidden, closed or private. AI usage should be explicit, configurable and limited by clear privacy rules. Since data may eventually move between the forum, Tornevall Tools and/or an external AI provider, we need to treat that data flow as something that requires careful handling.
We are therefore planning to keep building the AI integration with extra privacy and consent in mind.
The goal is to make AI assistance optional, visible and controllable. Administrators should be able to decide how much context is allowed, which provider is used, and whether AI features should be available at all in sensitive or private areas. But in the end, users should have the full right to control their data.
To address this, we are planning support for several privacy options (since this is a product for vBulletin, the plan is to make it generically controllable for anybody):
- Request only - only the user's own AI request is sent.
- Thread context - surrounding thread content may be included, if allowed.
- Consent required - content is only included from users who have explicitly allowed it.
- Opt-out mode - content may be included unless a user has opted out.
- Disabled - AI context handling can be turned off completely (on user level).
- Private node protection - private or restricted areas can force request-only mode.
We also want this to be configurable both globally in AdminCP and, where relevant, at profile level. A user's profile setting should be able to override the global setting when deciding whether their content may be included in AI context. It should also be possible for a user to disable AI context usage for their own content.
Another important point is provider visibility. Users and administrators should be able to see whether the configured AI route is:
- Tornevall Tools
- Direct OpenAI
- Another internal or custom provider in the future
This matters because consent to one provider does not automatically mean consent to another. Also, since we have funding issues, we should not handle ALL traffic from all forums in the entire world - that is a very stupid way to build this solution.
For closed, private or sensitive groups, the safest behavior should be the default: if the system cannot determine whether content may be used, it should not include that content. Quotes, signatures, attachments, embedded media, deleted posts, hidden posts and moderated content should not be sent as AI context by default.
The basic principle is simple:
A closed and safe group must remain closed and safe even when AI tools are enabled. External AI use must be explicit, explainable and easy to decline.
This feedback is useful and will be taken into account as the integration develops. The current implementation should be seen as an early prototype and proof of concept, not as a final architecture that every vBulletin installation is expected to use. The long-term direction is a more flexible setup where each forum administrator can choose the provider, data flow, consent model and security level that fits their community.
Follow the project at github, where you can also raise tickets for the product:
https://github.com/Tornevall/tools-vbulletin-ai
There is a generally understandable skepticism around the use of AI - not only from me, but from others as well.
A discussion in the vBulletin forum raised some important concerns about security, data protection and how forum content is handled when AI tools are involved. In our case, the current OpenAI integration is still an early prototype. We have not yet built the full privacy and consent layer around forum content, and that is exactly why this feedback is important.
The intention is not to silently send full forum discussions to an external AI provider, especially not when the forum are hidden, closed or private. AI usage should be explicit, configurable and limited by clear privacy rules. Since data may eventually move between the forum, Tornevall Tools and/or an external AI provider, we need to treat that data flow as something that requires careful handling.
We are therefore planning to keep building the AI integration with extra privacy and consent in mind.
The goal is to make AI assistance optional, visible and controllable. Administrators should be able to decide how much context is allowed, which provider is used, and whether AI features should be available at all in sensitive or private areas. But in the end, users should have the full right to control their data.
To address this, we are planning support for several privacy options (since this is a product for vBulletin, the plan is to make it generically controllable for anybody):
- Request only - only the user's own AI request is sent.
- Thread context - surrounding thread content may be included, if allowed.
- Consent required - content is only included from users who have explicitly allowed it.
- Opt-out mode - content may be included unless a user has opted out.
- Disabled - AI context handling can be turned off completely (on user level).
- Private node protection - private or restricted areas can force request-only mode.
We also want this to be configurable both globally in AdminCP and, where relevant, at profile level. A user's profile setting should be able to override the global setting when deciding whether their content may be included in AI context. It should also be possible for a user to disable AI context usage for their own content.
Another important point is provider visibility. Users and administrators should be able to see whether the configured AI route is:
- Tornevall Tools
- Direct OpenAI
- Another internal or custom provider in the future
This matters because consent to one provider does not automatically mean consent to another. Also, since we have funding issues, we should not handle ALL traffic from all forums in the entire world - that is a very stupid way to build this solution.
For closed, private or sensitive groups, the safest behavior should be the default: if the system cannot determine whether content may be used, it should not include that content. Quotes, signatures, attachments, embedded media, deleted posts, hidden posts and moderated content should not be sent as AI context by default.
The basic principle is simple:
A closed and safe group must remain closed and safe even when AI tools are enabled. External AI use must be explicit, explainable and easy to decline.
This feedback is useful and will be taken into account as the integration develops. The current implementation should be seen as an early prototype and proof of concept, not as a final architecture that every vBulletin installation is expected to use. The long-term direction is a more flexible setup where each forum administrator can choose the provider, data flow, consent model and security level that fits their community.
Follow the project at github, where you can also raise tickets for the product:
https://github.com/Tornevall/tools-vbulletin-ai
Comment