Officials fear that anti-government sentiment will undermine next year's census.
"Fake websites" are one of 16 key risks identified in the briefings released under the Official Information Act.
Briefings by Stats NZ to ministers show it is engaged in a $100,000 "national trust and confidence marketing campaign - focusing on myth-busting and managing disinformation about [the] census".
The department cannot afford another poor census: 2013 massively undercounted Māori, and 2018 tried to do the count mostly online resulting in Māori and Pasifika response rates plummeting.
The new threats are emerging several months after the Parliament occupation and protests.
"There is evidence of growing and significant anti-government sentiment and reducing trust and confidence in the government in parts of the community," a monthly update to ministers in August said.
In October, another update said this "continues".
"There is a risk that this negative sentiment will impact the trust and confidence in Stats NZ and/or the 2023 Census and will in addition negatively impact responses to the 2023 Census," it said.
It cost the government more than $9 million, and two years, to fix the 2018 census after Māori response rates fell to just 68 percent, jeopardising the usefulness of the census for steering health, education and other social spending.
Stats NZ in a statement told RNZ every census has some people objecting to it. "This census is no different."
Rising costs are another threat.
The entire census budget across five years is likely to hit $272m, according to a Stats report in September to the finance minister.
The official budget is $251m but this does not include a $14m Māori-led project or another $13m to "manage unknowns".
Two years ago Treasury was recommending $232m.
The reports show a $33m contingency is already tapped out. Postage and scanning is overspent by $11m and hiring temporary field workers was "coming in significantly higher than expected".
The government has trumpeted how it is doubling the number of field workers who go door to door, to 3500, to avoid the mistake of 2018, when they were cut to just 1700 as Stats tried to do much more online.
However, 3500 is still 1400 fewer field workers than Stats NZ said two years ago was the bare minimum.
It now says testing and analysis since then has changed the model. The 3500 would work "three times the hours worked in 2018" due to changed contracts.
The Stats report to the finance minister in September talks about a reduction: "To address the forecasted funding shortfall, the programme has identified cost control levers and a timeline, to reduce field quantities and volumes, if required," it said.
Another new and worrying risk around the Māori count appears in an update on 5 October.
The new $14m project, Te Mana Whakatipu, for iwi-led collection "has been progressing slower than expected, which has started impacting collection operations development work. There may be a need for an alternate approach to collections, which may impact response rates for Māori".
However, last night the deputy government statistician Simon Mason said "this isn't technically correct".
Te Mana Whakatipu was both "later than expected" and "on track".
"Te Mana Whakatipu will have no impact on response rates or collection approaches for the rest of the country," he said, without explaining why the briefing said the opposite.
A Māori response rate under 90 percent would be widely seen as a failure.
Stats has admitted trust in the whole process was undermined by the 2018 debacle.
"Census teams are working closely with communities to rebuild trust in census," it says.
Regards anti-government sentiment, "marketing, communications, and engagement activities are specifically focused on building trust and confidence ahead of the census."
Those activities are costing around $6m.
Local body elections turnout bad sign
Stats was also aiming to take lessons from last month's local body elections, when voter turnout slumped to a record low 36 percent nationwide.
The department has been trying to reassure people online, talking about "multiple layers of security", that everyone who accesses the data signs a lifelong declaration of secrecy, and that all the data is stored by the government in the cloud - this was in servers onshore, Stats said.
Stats is also getting an independent review of its approach to privacy.
In 2018, fully 700,000 people did not do the census. However, convictions and fines for being a census refusenik are very low, though rising, from 41 in 2006 to 60 in 2018.
The census date is 7 March.
Inflation, IT problems and the hangover of Covid are other obstacles, the OIA report show.
Stats NZ is having to spend almost $1m fixing its statistical computing environment because of "stability, performance, security and license issues deeming it not fit for purpose to meet census use cases".
The final $17m of backup contingency money was released by the government a month ago.
Stats asked for it, arguing census 2023 had started on the backfoot, due to clean-up work on the botched 2018 census taking till mid-2020, and citing rising costs.
"Field staff roles are more difficult to recruit for than originally anticipated," one report said.
Cabinet agreed earlier this year to. amend the law to let census take place in 2024. However, the government statistician said 2023 would be OK.
Stats got an external report done that estimated the benefits of a census at $2.8 billion.
Covid forced big changes earlier this year to a planned test of an intensive fieldwork approach in South Auckland and Bay of Plenty.