ChatGPT Team - are chats really not used to train models?

ChatGPT
1

Interesting observation in dev tools console:

image
image1906Ɨ640 133 KB

My cocncern is - are our data really safe whe we are using Team Plan?

3 Likes
2

I donā€™t see that display in my environment, but just to be sure, could you please check if ā€œChat trainingā€ is turned ON and let me know?

3 Likes
3

In case you were curious, this setting is still sent even when ChatGPT client is in ā€œno historyā€ mode.

2 Likes
4

In Team plan itā€™s impossible to turned it on :wink:

image
image1276Ɨ796 41.9 KB

1 Like
5

Does it mean, that only a setting is on, but data is safe? I am trying to understand what OAI is doing :wink: And this is critical if there is some ā€˜bugā€™, which causes the query data in the Team plan and in"no history" mode, the data is still used by OpenAI for trainingā€¦

1 Like
6

Just in case, I thought it would be better to handle this issue as a problem on Bugcrowd rather than asking directly in the help section, so I reported it with the following two possibilities in mind:

  • The response data only indicates that ā€œtraining is allowedā€, but in reality itā€™s not being used as training data.
  • As the response data indicates, itā€™s actually being used as training data.

Within a few hours, I received a response stating that it ā€œdoes not pose a security risk to the platformā€ and that the vulnerability was ā€œnot applicableā€.

To be sure, I followed up with a question asking for confirmation that conversation histories are not used as training data, but there was no additional response on Bugcrowd (maybe I asked in the wrong place).

As the ā€œNot Applicableā€ designation indicates, this is not considered confidential, so I am sharing my experience here for your reference.

The following two points need to be considered:

  • The ChatGPT Team service is subject to the same ā€œBusiness Termsā€ and " Enterprise Privacy Policy" as the ā€œOpenAI APIā€ and ā€œChatGPT Enterpriseā€.
  • If itā€™s actually used as training data, it would be a serious negligence on the part of OpenAI.

I think an explanation from OpenAI is urgently needed, but personally I am not too worried about it.

For those who are concerned, I recommend not to use confidential information with the ChatGPT Team until an explanation is provided by OpenAI.

If I ask a question to OpenAI help and get an answer, I will let you know.

1 Like
7

Thank You for your engagement and all the information.

Yeah, I hope that this is exactly this case:

Concerning:

I know trust portal: https://trust.openai.com/ and I hope that there isnā€™t any security breach in this situation.
I thought that maybe here someone from OpenAI will answer, but I see that have to contact with support.

1 Like