AI and automated tools
We use computer programs to make reporting quicker. This page explains what they do, what they see, and what people decide.
The short version
- Programs make suggestions. You check them before you submit, and a caseworker reviews every report.
- Screenshot reading happens on your own device by default.
- Programs do not use your name, email or phone number.
- No decision that significantly affects you is made by a program alone.
- Your reports are not used to train AI models.
What the tools do
Reading screenshots (quick report)
When you add a screenshot to quick report, a text reader in your browser reads the words in the image. Your browser downloads the reader from a public code library service (jsDelivr), but your screenshot stays on your device. We use the text to suggest the platform, account name, date, links and why it may be anti-Hindu. Suggested answers are marked Suggested until you check or change them. Nothing is sent until you press submit.
Suggesting a category
As you answer questions, a set of written rules on our server suggests what kind of incident it might be and which signs of anti-Hindu bias are present. Your answers are sent to our server in Sydney for this, but are not saved until you submit. You can accept or change the suggestion.
Optional AI reading
We have built an option to use an AI model from Anthropic to read screenshots and incident text more accurately, including posts in other languages. It is switched off. Before we switch it on we will update this page and our privacy policy, and tell you on screen when AI reading is in use. If it is used, only the screenshots and incident text you choose are sent, never your contact details, and Anthropic may not use them to train its models.
What people decide
Caseworkers decide whether a report is in scope, how it is classified, whether it is verified, whether it counts in our figures, and whether it is shared or referred. Suggestions from the tools are recorded as suggestions so caseworkers can see where an answer came from.
Information used by these programs
The incident description and answers, links, account names and text read from screenshots. The programs make or help make suggestions about the kind of incident and the signs of bias. They do not make decisions about you, your rights or your access to support.
Limits
Text readers make mistakes, especially with small, stylised or non-English text, and rules can misjudge context such as sarcasm. That is why every suggestion can be changed and every report is read by a person. Tell us at ggupta@saraswati.care if a suggestion was wrong or offensive, so we can improve the rules.