A federal parliamentary inquiry is currently asking a difficult question: what is driving racism, hate and violence against Aboriginal and Torres Strait Islander people, and what role does the internet play?
The inquiry follows several violent incidents, including the attack on Camp Sovereignty in Melbourne in August 2025 (and importantly, a new attack on August 2 2026), and the alleged attempted bombing of an Invasion Day rally in Perth.
At public hearings around the country, First Nations people and organisations have described racism in workplaces, schools, public institutions and online. They have also warned racist abuse is becoming more open, more organised and easier to spread.
New research helps explain the patterns behind those experiences. By understanding how this online hate works, we can better stop it.
The research
Researchers at the Tackling Hate Lab analysed more than three million posts and news items about the 2023 Voice referendum, Anzac Day 2025 (when a group of men were charged for heckling during a Welcome to Country) and the so-called March for Australia rallies in 2025. The data came from X, news media, Reddit, YouTube, Facebook, forums and other online platforms.
The study used captured hateful language using Google’s Perspective API, an artificial intelligence (AI) model.
We also used a bespoke AI model trained using judgements from First Nations community annotators to identify anti-First Nations hate. This matters because deciding what counts as racism cannot be separated from the experiences of the people being targeted.
Hate changes with the conversation
Political events reshape how hate targeting First Nations is expressed. Public debates and media moments make it louder and change the dominating narratives.
During the Voice campaign, anti-First Nations hate increased in stages. At the beginning of 2023, the study system detected between 200 and 400 hateful posts on most days.
By September, daily numbers were regularly above 2,000. In early October, immediately before the referendum, they reached about 3,600 posts per day.
The pattern looked less like a single explosion and more like a staircase. Each major campaign moment (such as the launch of campaign advertising, for example) increased hostility.
This suggests political campaigns don’t just reflect existing prejudice. They can create moments when hostile ideas become easier to express, repeat and share.
First Nations public figures received much higher levels of racialised abuse than many non-Indigenous public figures with comparable followings on social media. More than a quarter of all posts mentioning then-Indigenous Australians Minister Linda Burney in the dataset were classified as racial identity hate. Lidia Thorpe, Noel Pearson and Adam Briggs were also targeted at high rates.
Racism doesn’t always look like a slur
A second finding is that anti-First Nations hate changes form depending on the context.
During the Voice campaign, almost 89% of the hate detected focused on the referendum itself.
Importantly, the study did not classify all opposition to the Voice as hate. Statements such as “Vote No”, or First Nations critiques of the referendum, were not coded as hateful.
Instead, First Nations community annotators identified as hateful language that which portrayed the Voice as a form of racial oppression of non-Indigenous people, or denied the historical and ongoing effects of colonisation and racism.
In these cases, the concern was not disagreement with a policy proposal, but language that community members considered to be hateful because it denied First Nations’ experiences or framed their recognition as a threat to other Australians.
