Regression: "textColumn no like 'xy'" is now "invalid where clause"

For years we used the following query:

IsDeleted__c!=true AND (IsPublished__c = true and (Unrestricted__c = true OR Restricted_Buyers__c not like '%C22FC2B7-2C7E-1B22-FF04-8FFB6EAD2000%'))

notice that Restricted_Buyers_c is a TEXT field.
Leads to the following error:

{"code":1017,"message":"Invalid where clause at position 102: mismatched input 'like' expecting IN"}

Application: 3EC9156B-FDDA-1998-FF35-4DA70D8C2800
We will now try to work around this issue, yet this has to be addressed.

It is however possible to use the following workaround:

NOT(Restricted_Buyers__c like 'STRING')

Hi Mathis,

Thank you for reporting - this is really a regression. An internal task BKNDLSS-13439 has been created to resolve it.
For now, we’re reverting the latest release now so that the old query would continue to work as earlier. We shall notify you as soon as it happens.

I would appreciate if you add a test not only for “not like”, but also “NOT(… like…)”, just in case this happens again in the future.

Yes, sure, we’ll do our best to support the new format, too.

The old query works now.