A convincing job offer, a message that mentions where you work or a caller who sounds like a relative could all be part of scams supported by AI agents, cybersecurity experts have warned.
Criminals are increasingly using AI agents to research targets, handle conversations and carry out different parts of attacks, reducing work that previously required more people and technical knowledge.
AI agents can make an approach feel personal, answer questions, and keep trying when the first request fails, which makes them increasingly dangerous for potential victims.
“Agents are now doing the work, not just advising,” Farid Zahran, senior managing director of AI and digital transformation at FTI Consulting, told The National.
With the help of other tools, AI agents can gather information, send messages and follow up with potential victims. The ability of AI agents to carry out elaborate tasks is making it harder for victims to spot potential scams than it was when human error was a sign.
Premchand Kurup, chief executive of ParamountAssure, said this made some scams easier and cheaper to run.
“Tasks that once required technical expertise, time and a team can now be carried out with fewer resources.”
But using AI did not mean that every scam was being independently run by a machine; humans are still involved to set goals, oversee attacks and control the collection of stolen money, experts cautioned.
A scam that knows more about you
Scammers can use AI to search social media, professional profiles and company websites before making first contact.
They could use information about a recent holiday, a new job or an employer to write a message that seems believable.
“A scam no longer needs to rely on a poorly written message sent to thousands of people. AI can make fraudulent communications more convincing and help criminals tailor their approach to individual targets,” said Vladislav Tushkanov, group manager at Kaspersky’s AI Technology Research Centre.
That could be in the form of a fake recruiter who knows someone’s work history or a fraudster posing as a familiar contact.
AI can help process the information quickly, but it does not automatically give criminals access to private messages or bank records, Mr Tushkanov said.
Keeping the conversation going
Unlike a fixed script, an AI agent can respond to what a person says and store memory of earlier conversations.
If someone questions a payment request, the agent can change its approach or even offer an explanation to allay doubts among potential victims.
“An agent can change its reply based on what someone says and remember what has already been discussed,” Mr Kurup said. “This can make the exchange feel like a natural conversation rather than a set of fixed replies.”
That could help with scams that depend on building trust, including romance fraud, fake job offers and investment schemes.
Agents can also manage many conversations at once and keep separate records for each person. But a human may still need to step in when a target becomes increasingly suspicious or the conversation takes an unexpected turn.
“The ability to generate convincing content doesn’t mean that an agent can reliably impersonate a person or complete every stage of a scam without supervision,” Mr Tushkanov said.
Attacks are already happening
In an August 2020 report, Anthropic said an attacker had used its Claude Code tool in an extortion operation against at least 17 organisations, including healthcare providers, emergency services and government bodies.
The tool helped break into networks, analyse stolen information and prepare demands for money. Anthropic said it disrupted the activity.
The AI helped assess stolen financial information to determine how much to demand. Some ransom requests exceeded $500,000, Mr Zahran said.
The same report described another criminal with basic coding skills who used AI to develop ransomware, software to lock or encrypt files and to demand payment. Packages were offered to other criminals for between $400 and $1,200.
A September 2026 Anthropic report described more cases involving AI helping carry out attacks and steal data. Human operators still selected targets and reviewed the stolen information.
The UAE is ranked sixth internationally as a target of cyber attacks in an analysis by Microsoft.
The US topped the rankings in Microsoft's Digital Defence Report, which monitored cyber crime activity affecting customers from January to June of this year.
Next came Israel, Ukraine, Taiwan, the UK and the UAE. India, Japan, Canada, Germany, Australia and the Philippines completed the top 12.
What should make you suspicious?
Bad spelling and awkward wording are no longer reliable warning signs.
“AI can write convincing messages that include details about the recipient. A familiar voice or video is not proof of identity either, because both can be faked,” Mr Kurup said.
People are advised to look closely at requests. Odd signs include unexpected demands for money, pressure to act quickly, changed payment details or requests for passwords and verification codes.
Even if a message is from a familiar account, it is still advised to check before responding.
Mr Tushkanov advised checking unusual requests through a known telephone number, an official app or an established contact.
“Avoid using contact details or links supplied in the suspicious message itself.”
If you have already clicked on a link or transferred money, contact the bank or payment provider immediately to check whether the payment can be stopped or recalled. Then report the incident to the police.
Keep proof of messages, screenshots and payment references. If a password has been shared, change it on a trusted device and update any other account where it was reused.
Those who have downloaded a suspicious file or have installed software should seek technical help, particularly if someone has gained access to the device.
Mr Tushkanov warned against continuing the conversation in an attempt to get money back.
“Early action may improve the chances of limiting further losses, although recovery cannot be guaranteed.”
