If you thought drive-locking was wrong...

[Telos]

With the current release of Synology Office, users are now required to install Synology's AI product on their NAS (will Synology Drive, and others be next).

From the release notes:

How special. We have the grand opportunity to train Synology's AI on our personal data.

What could possibly go wrong?

Synology's death knell is ringing louder.

 

[White Bear]

Interesting... Looking at Synology's statements on Synology Office Suite meets AI it seems that this is marketed as an AI integration tool that can connect to OpenAI, Azure OpenAI, Amazon Bedrock, Google Gemini, Baidu AI Cloud, etc. (vs. a standalone AI).

HOWEVER, there's a paragraph about privacy that seems a bit ambiguous. Synology says that our data "remains safe from training by third party AI". But says nothing about data being safe from Synology. Is that your concern?

Your data remains securely on your Synology NAS, ensuring your business information remains safe from training by third party AI providers.

 

[Telos]

Is that your concern?

There's that, as well as all the vulnerabilities that arise through Synology shoddy coding QC.

AI should be entirely optional. Instead, it is heading towards Synology Application Services.

What could possibly go wrong?

 

[Rusty]

While the Console does install you don’t have to configure it with AI providers. Also if you do, you still have to activate the usage of said provider via specific productivity app. So it’s not automatic by any means.

 

[Telos]

you don’t have to configure it with AI providers

Odd then, that installation is mandatory ... and for the privacy-minded, the installation of Container Manager is also required.
Perhaps Synology has cards up its sleeve that have not yet been revealed.

 

[Rusty]

Odd then, that installation is mandatory ... and for the privacy-minded, the installation of Container Manager is also required.
Perhaps Synology has cards up its sleeve that have not yet been revealed.

I would agree.

 

[White Bear]

More privacy information from Synology here: SYNOLOGY AI SECURITY & PRIVACY PRACTICES. Looks like there are significant privacy vulnerabilities in the "De-Identification" process. Glad I don't have a need for AI integration.

 

[fredbert]

The de-identification feature is to stop passing known patterns from being sent to the various third party AI services. This hase variously been known as data loss protection (DLP) and applied in next-gen firewalls, cloud access security broker (CASB) services, and so on which are now fashionably covered under security service edge (SSE) or the fuller secure access service edge (SASE).

Sure this can be done on-box, and I'm not sure why it isn't unless it's to do with NAS performance: these can take a lot of resources. There is also the ability to improve the pattern matching centrally without the need to patch the DSM AI package.

I still agree that it shouldn't be a mandatory package just to use Synology Office. If I wanted Office to use AI then I should be able to install AI Console then. Seems that a few conditional if statements could have sorted this out.

Ultimately though what it's doing is to stop your identifiable data being sent to AI services, but it is still going out to Synology. Which would you trust more? Or you can just be a little smarter and not include your identifiable data in submissions to AI services. The problem comes in getting other less-aware users to pre-screen their submissions... your know the ones: they have open social media accounts and post when they go on holiday... please burgle my house!!!

 

[Rusty]

unless it's to do with NAS performance: these can take a lot of resources.

hate those... DigitalGuardian anyone -.-?

 

[Akount]

good point Rusty, there are other DLP providers if we don't want to trust Synology with all AI i/o calls (personally I don't have a reason to distrust synology as long as they give us options). from my perspective, the de-identification thing if disabled, means that there is no real privacy risk since they are not able to monitor the API requests it's going straight to the model providers. since AI-enabled office is now out of beta I am glad it has the ai stuff (I posted about last month analyzing how unclear the beta packages' privacy terms are). This is the only real synology forum right (by far and away the best, most active), and that was the only ai office post here, I wonder if they saw that and were like 'f*ck lets just ship it to non-beta now' because this looks rushed, like it says you can monitor all inputs and outputs but in the logs it just shows token count, does not display the actual inputs or outputs. Also, the models are not yet selectable.. it gives us only the low-power models and no option for self hosting which most self-hosters probably use LMStudio, and most businesses are now able to spin up a workstation or some other endpoint with GPU's pretty easily (ever heard of nvidia's DGX spark coming next month)? It no longer costs 6-7 figures to fine tune a model and deploy it on your own hardware, depending on details this should cost 4-5 figures total so all businesses will be doing it and they will not want to use the low-power models currently available thru the package. Even if you don't self host you still need the best models (gpt-5 , 2.5 pro, etc, if it's a serious/business application I believe). I am a big supporter of what Synology is doing with the new google drive alternative but if we are going to benefit from the ai package it has to be way better, they must let us upload docs & system prompts and call it an "agent" that is easy to do over the api's. I mean this is 2 years old when chatGPT could make your own GPT's and ms365 + google workspace has these throughout their ecosystems.

 

[Jan Janowski]

Though I don’t presently use Synology Office, (Presently using the free version of WPS Office, only need it for one person, me), and selected it because it displayed Word files with graphic boxes the most accurately on small iPhone display without scrambling. (Did not try SO, by the way).

AI: I’m hoping that after some amount of time AI can/could/hopefully-will be beneficial….. and not a trap, as it reportedly seems or has inadvertently has or may have become - in many developer’s hands. It’s a bit unclear, and I’d rather react on the side of caution for now.
Too early - Too potentially buggy: in my opinion - to trust right now.
Time will tell, but for now: tested by others, not me. But I’m just a self taught know-nothing……. But has good “gut feelings”

 

[Rusty]

Also, the models are not yet selectable.. it gives us only the low-power models and no option for self hosting which most self-hosters

I’m testing a few new productivity apps that will soon ship as well as the new 1.2 version of the Console. It offers self hosting as well as for example gpt5 and 5mini.

 

[Telos]

no option for self hosting

Sending your content off-site for AI processing is a recipe for peril and privacy compromise. This is quite different from asking ChatGPT where to vacation, as in that scenario, the user determines the shared content. Further, requiring the AI package as an integrated service, is simply wrong. Just another reason which weakens one's ongoing relationship with Synology.

 

[fredbert]

If all you are using AI for is generic questions where you provide little information then there is no need for DLP/CASB. But seems there is more focused use, and when coming from an office document which can have personal / corporate data, then it could be wise to screen what is leaving your control. Also factor that persistent dialogues can now happen, that could leave a lot of data exposed or harvestable.

So you decide you use a trusted third party DLP/CASB service that monitors and intervenes on user activity, according to set policies. In this case Synology Office is linking to Synology’s AI screening service to perform this for you. Sure there could be alternative services being offered: in my experience CASB services have been offered by separate vendors, while DLP is often part of the security control services you use. Though now single vendor offerings combine all these features: NG FW; AV/ABot; URL; App Controls; Sandbox; DLP; CASB; SD-WAN; etc.

For business customers of Synology they may place Synology in the trusted vendor space, and so trust them to screen AI communications. That’s if they use Synology Office: most I see use M365 or Google.

 

[Akount]

Sending your content off-site for AI processing is a recipe for peril and privacy compromise. This is quite different from asking ChatGPT where to vacation, as in that scenario, the user determines the shared content. Further, requiring the AI package as an integrated service, is simply wrong. Just another reason which weakens one's ongoing relationship with Synology.

I agree with you. It's so easy to grab a local model endpoint, LM studio makes this so even novices have this option. Other ai tools I have let me plug in a local end point from lmstudio or ollama, Synology should do that since it's all about self hosting.

 

[Jan Janowski]

Wanting to know more — How is AI Useful for me? — I watched part of NAS COMPARES’s video on AI and Office, and gave up during the demonstration of how it creates a negative response to something as an email reply…..
Are You Kidding? This is progress?
Last thing I’d want: An automatic: ‘BS Generator’! ????
Instead of paragraphs of flowing prose……. a simple ‘No’ would suffice!
So is AI - a ‘BS Generator’ for those afraid to just say ‘No’ ????
Now, just imagine this reply being a few ten’s of thousands of words long…….

 

[Rusty]

Synology should do that since it's all about self hosting

AI Console 1.2 offers selfhosting

 

[fredbert]

Wanting to know more — How is AI Useful for me? — I watched part of NAS COMPARES’s video on AI and Office, and gave up during the demonstration of how it creates a negative response to something as an email reply…..
Are You Kidding? This is progress?
Last thing I’d want: An automatic: ‘BS Generator’! ????
Instead of paragraphs of flowing prose……. a simple ‘No’ would suffice!
So is AI - a ‘BS Generator’ for those afraid to just say ‘No’ ????
Now, just imagine this reply being a few ten’s of thousands of words long…….

The merits of AI, or not, has been done to death cross most media outlets for at least a couple of years.

What we’re discussing is Synology’s integration, in particular sending your interactions to AI to Synology to screen out identifiable data. It isn’t really uniquely a problem with AI, rather the problem arises from sending data to an intermediary screening service that you must trust.

A previous use of this type of screening would to ensure business data isn’t being emailed out. The policies would be adapted so that some corporate user groups are able send or export such data but most employees are not.

 

[Akount]

AI Console 1.2 offers selfhosting

That is super exciting! How do you know? 1.2 isn't shown as an option past v 1.1 on my package center even if I enable to view beta packages (which after confirming beta packages privacy policy is like most, they broadly disclaim the privacy protections of the main Synology privacy policy and even advise no use in commerce), I stopped considering them.
btw did anyone see the new 96 gb vram ddr4 gfx card announcement from huawei?, I personally wouldn't be able to use Chinese hardware since they don't seem to have a court system which would dissuade Leadership from exploiting backdoors, but maybe the AMD cards (also coming q4) which are 48gb for $1250 will suffice (they are ddr6).

 

[Rusty]

How do you know?

Because I have that version in for testing.