Hackers using AI-produced audio to impersonate tax preparers, IRS

Artificial Intelligence has supercharged an array of tax-season scams this year, with fraudsters using deepfake audio and other techniques to intercept funds and trick taxpayers into sending them financial documents.

Cybercriminals have long used tax season lures during the first four months of the year, but multiple cybersecurity experts have recently pointed to one worrying update to the schemes: AI-enabled voice phishing attacks.

Hackers are using AI-generated audio to pretend to be a person’s tax preparer, accountant or the IRS — using previously stolen personal information to lend credence to the scam.

Bugcrowd founder Casey Ellis said generative AI and deepfakes are “game-changers for scammers.” 

“They allow attackers to scale their operations while increasing the believability of their scams,” he said. “For example, a deepfake video of a ‘tax advisor’ could be used to lure victims into sharing sensitive information, or AI-generated emails could mimic the tone and style of legitimate IRS communications with uncanny accuracy.”

He said they have seen a sharp increase in AI-driven attacks that help cybercriminals create “highly convincing phishing emails, voice calls and even video messages.”

Deepwatch CISO Chad Cragle and several others separately confirmed that they are seeing an increase in AI-driven attacks this tax season, particularly with convincing phishing emails and deepfake audio designed to impersonate trusted tax officials. 

Cragle added that attackers are replicating voices to deceive victims over the phone and are offering to help victims create IRS online accounts, prompting them to hand over sensitive financial information. 

Patrick Tiquet, a vice president at Keeper Security, noted that cybercriminals can now create realistic video and audio impersonations of IRS agents, tax professionals or even family members, tricking individuals into divulging information like Social Security numbers or tax credentials. 

Tiquet and Ellis said people should look out for inconsistencies, noting that AI often struggles with the finer details. People should also verify identities before handing over information and refuse any demands that sound urgent, either over the phone or in emails. 

Some reverse image and video search tools also will notify you if content has been artificially generated or manipulated, Ellis added. 

Texts, typosquatting and Trump refunds

In addition to AI voice scams, defenders continue to see novel variations on tried and true tax season cyber theft schemes.

Zimperium’s Kern Smith said he has seen a rise in mobile-first attacks, where attackers send texts posing as the IRS or a tax service, urging recipients to click on malicious links or download fake apps. 

These scams are designed to steal login credentials, Social Security numbers and other sensitive financial information, Smith said. 

Experts also spotlighted a panoply of phishing sites and spoofed platforms designed to take advantage of people searching for terms like “Trump tax refund” or others looking for well–known tax companies like H&R Block. 

Cragle said counterfeit sites are leveraging SEO poisoning, altering search engine rankings to seem legitimate and lure in victims while Ellis added that cybercriminals are also exploiting unpatched vulnerabilities in tax software or third-party integrations.

Microsoft published a lengthy blog last week highlighting recent phishing campaigns that spread malware through emails with subject lines saying things like “Notice: IRS Has Flagged Issues with Your Tax Filing,” or “Important Action Required: IRS Audit.”

Many of the emails had PDF attachments with IRS-related file names and targeted both customers as well as CPAs and accountants. 

“Between February 12 and 28, 2025, tax-themed phishing emails were sent to over 2,300 organizations, mostly in the United States in the engineering, IT, and consulting sectors,” Microsoft said, noting that the emails were empty but had PDF attachments with QR codes linking to malware.