AI makes it easier to tailor scams to individuals, but the everyday defences are refreshingly concrete: use two-step verification, keep important systems updated, and check requests for money through a number you already trust. An opinion article by Tsinghua University professor Duan Haixin also warns that AI agents with access to email, files or accounts need tightly limited permissions.
Watch Desk analysis
What happened
Duan’s article for CGTN argues that AI can help scammers personalise messages at scale, including by imitating familiar voices or identities. It also highlights the risk of giving an AI agent more access than a task requires. These are the article’s assessments, not independently established findings about the prevalence of particular scams.
For readers, Duan’s practical recommendations include using different passwords for important accounts, turning on two-step verification, installing software and AI services only from official sources, and keeping an offline backup of important files. For requests involving money or identity checks, verify through a number you saved yourself or an official channel, not a voice or video alone.
Why it matters
The advice is useful because it treats AI as an amplifier of familiar risks, not as a reason to panic at every odd email. The agent point matters too: if software can act on your accounts, limiting what it can access also limits what a hijacked or misdirected agent could do.
Duan’s wider argument is that personal vigilance cannot do all the work. Companies and product makers have responsibilities for safeguards and prompt security fixes, while public policy has to address the systems around a model, not just the model itself.
Our read
Start with the unglamorous basics: two-step verification, offline backups and a second channel for checking urgent requests. They are not a force field, but they do make a scammer’s job less convenient. And if an AI agent does not need access to your inbox or payment account, do not give it the keys just because a permission screen looks friendly.
What to watch
- Whether AI services make it easier to review and restrict an agent’s permissions.
- Whether companies give users clearer ways to report scams and suspected account misuse.
- Whether policy proposals translate into practical, enforceable protections for users.
Discussion spark: Should AI agents be allowed broad access by default for convenience, or should every permission need to be granted task by task?
Sources and evidence
- Who will protect ordinary users as AI fuels cyberattacks? (17 November 2026, 12:10 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.