I wouldn't immediately conclude that the service provider had manipulated the data. Theoretically, duplicate respondent numbers in CAPI shouldn't happen very often. But, it turns out that the random number generator we use for that particular feature isn't nearly as random as we'd like, and duplicates are actually pretty common. (Note that the random number generator we use for CAPI is NOT the same one we use for most other processes, such as conjoint design generation.)
On a recent survey with about 600 respondents, we found that duplicate respondent numbers had occurred about 30 times. And, the more respondents you have, the more likely it is that duplicates will occur, even if you had access to a perfect random number generator! If we used a perfectly random generator to pick 100 numbers between 1 and 1,000,000, there's a 0.5% chance that we'd draw at least one duplicate. With 200 draws, the cumulative probability of a duplicate jumps to 2%. With 500 draws, there's a whopping 12.5% chance of a duplicate, and with 1,000 draws, there's about a 50/50 chance that at least one duplicate exists.
More recent versions of SSI Web (starting with SSI Web v8.2.0) implement a smarter CAPI data merge, that reassigns respondent numbers when duplicates are encountered. It stores the original number as well, so it can recognize the duplicate later if it encounters it again.
So, while I wouldn't rule out data tampering, this certainly isn't a "smoking gun" indicating that data manipulation has occurred.