Reuters has published a look at the problems the Census Bureau is facing in its attempt to hastily finish software and patch problems ahead of next year's mammoth official national census. As usual, the government's projected costs for the new systems, which will make the 2020 census the first truly digitized version, have "doubled," to $167 million. As usual, those cost overruns come from a leadership insistence on hiring outside contracting firms rather than building the software the census needs using government workers—an especially odd move given the census' status as a unique undertaking that cannot reasonably be approximated using off-the-shelf commercial products.
But the most dire concerns are that the nation's first mainly online census might not work at all. Reuters reports that a large-scale system test in 2018 was both successfully hacked by unknown Russian attackers and subjected to a denial-of-service attack that the system's primary outside security contractor was unable to properly prevent or even satisfactorily diagnose. The test took place against internal government technology experts' advice that security issues be resolved beforehand, and it's not clear that large portions of the software will be finished, or tested, or hardened against future hackers before the census gets underway.
And in yet another parallel to countless corporate technology efforts to save money by hiring outside firms to handle core technology needs, the Census Bureau has tasked its own original programming team with finishing out the internal software it originally developed just in case the contracted version isn't ready in time or goes belly-up when census time actually arrives. (When Republicans threatened to bring a big-business management approach to government, they were not messing around.)
It is the security concerns that are most significant. Critics are raising several issues: one, that foreign attacks may render the system overloaded or nonoperational for the public or for census employees themselves. Two, that the data could be stolen. Three, and the one suggested by the successful Russian incursion into the test system: that the data could be altered, perhaps in ways that sabotage the very intent of the census. The last one of those seems extraordinary; it would take a monumental blunder to allow it, but the attractiveness of census data to the world's most sophisticated state-sponsored actors cannot be overstated.
All that said, while the Reuters report is an important look at what's going wrong in the census efforts, it's difficult to parse out just how threatened the census itself may or may not be. It is not unusual for megascale technology projects to have steep cost overruns. It is somewhat alarming, but commonplace, for technology teams to construct insecure systems with the intent of fixing them much later in the process. And it isn't unusual to have both outside consultants and inside employees giving dire warnings about a project's flaws to those willing to listen: If things do go belly-up, there is an advantage to being the voice that told management, I told you so, or, if all goes well, self-identifying as the voice that made everyone focus on the problems before it was too late.
It's just another thing to worry about as 2020 shapes up to be one of the more monumental years in modern American history. And yes, by monumental I mean potentially very very bad.