AI Is Making South African Tax Scams Harder to Recognise

Artificial intelligence is removing some of the obvious warning signs that once exposed tax scams in South Africa. As the 23 October filing deadline approaches, fraudsters are using generative tools to produce polished messages that imitate the language and presentation of the South African Revenue Service.

Kaspersky researchers identified a fake SARS eFiling page designed to collect usernames, passwords and telephone numbers. A second campaign uses a counterfeit Adobe PDF sign-in screen to capture email credentials. The messages are built around familiar tax-season triggers, including refund promises, assessment notices and threats of legal action.

The change is important because many people have been trained to look for poor spelling, awkward formatting or amateur design when judging whether a message is genuine. Those signals are becoming less reliable. A message can look professional and still be a phishing attempt, particularly when it directs a recipient to a site that copies the branding and layout of an official service.

Kaspersky also warned about a different risk: taxpayers turning to public AI tools for help with returns and tax questions. Chatbots can explain general concepts, but users should not upload tax records, identity documents, banking details or other confidential information. The data may be retained or exposed through sharing features, and the answers may contain errors that are difficult to detect.

The practical response is straightforward. Taxpayers should type the official SARS address into a browser rather than follow links in an email or message. Refund offers and summons threats should be treated with suspicion, and eFiling credentials should never be entered on a page reached through an unsolicited link. When in doubt, the safest route is to verify the message through SARS’s official contact channels.

The warning also fits a wider African cybersecurity pattern. AI is lowering the cost of producing convincing fraud, allowing campaigns to scale faster and making it harder for users to rely on appearance alone. Organisations will need stronger authentication, public education and secure reporting channels alongside technical controls.

For citizens, the central lesson is simple: visual polish is no longer evidence of authenticity. The source, the address and the verification process matter more than how convincing a message looks.