This is according to cybersecurity and digital privacy company Kaspersky, which said it detected and blocked over 142 million phishing link clicks in Q2 2025, a 3.3% increase globally and a 25.7% increase in Africa from Q1.

AI has elevated phishing into a highly personalised threat, Kaspersky said. Large language models (LLMs) enable attackers to craft convincing emails, messages and websites that mimic legitimate sources, eliminating grammatical errors that once exposed scams. AI-driven bots on social media and messaging apps impersonate real users, engaging victims in prolonged conversations to build trust. These bots often fuel romantic or investment scams, luring victims into fake opportunities with AI-generated audio messages or deepfake videos.

Attackers also create realistic audio and video deepfake impersonations of trusted figures — colleagues, celebrities or even bank officials — to promote fake giveaways or extract sensitive information. For instance, automated calls mimicking bank security teams use AI-generated voices to trick users into sharing two-factor authentication (2FA) codes, enabling account access or fraudulent transactions. Additionally, AI-powered tools analyse public data from social media or corporate websites to launch targeted attacks, such as HR-themed emails or fake calls referencing personal details.

Employing new tactics to bypass detection

Phishers are deploying sophisticated methods to gain trust, exploiting legitimate services to prolong their campaigns. For instance, Telegram’s Telegraph platform, a tool to publish long texts, is used to host phishing content. Google Translate’s page translation feature generates links that look like https://site-to-translate-com.translate.goog/… and are used by attackers to bypass security solutions’ filters.

Attackers now also integrate CAPTCHA, a common anti-bot mechanism, into phishing sites before directing users to the malicious page itself. By using CAPTCHA, these fraudulent pages deflect anti-phishing algorithms, as the presence of CAPTCHA is often associated with trusted platforms, lowering the likelihood of detection.

A switch in hunting: From logins and passwords to biometrics and signatures

The focus has shifted from passwords to immutable data. Attackers target biometric data through fraudulent sites that request smartphone camera access under pretexts like account verification, capturing facial or other biometric identifiers that cannot be changed. These are used for unauthorised access to sensitive accounts or sold on the dark web. Similarly, electronic and handwritten signatures, critical for legal and financial transactions, are stolen via phishing campaigns impersonating platforms like DocuSign or prompting users to upload signatures to fraudulent sites, posing significant reputational and financial risks to businesses.

“The convergence of AI and evasive tactics has turned phishing into a near-native mimic of legitimate communication, challenging even the most vigilant users. Attackers are no longer satisfied with stealing passwords — they’re targeting biometric data, electronic and handwritten signatures, potentially creating devastating, long-term consequences. By exploiting trusted platforms like Telegram and Google Translate, and co-opting tools like CAPTCHA, attackers are outpacing traditional defences. Users must stay increasingly sceptical and proactive to avoid falling victim,” said Olga Altukhova, security expert at Kaspersky.

Earlier in 2025 Kaspersky detected a sophisticated targeted phishing campaign which was dubbed Operation ForumTroll, as attackers sent personalised phishing emails inviting recipients to the “Primakov Readings” forum. These lures targeted media outlets, educational institutions and government organisations in Russia. After clicking on the link in the email, no additional action was needed to compromise their systems: the exploit leveraged a previously unknown vulnerability in the latest version of Google Chrome. The malicious links were extremely short-lived to evade detection and in most cases ultimately redirected to the legitimate website for “Primakov Readings” once the exploit was taken down.

To be protected from phishing, Kaspersky recommends:

Verify unsolicited messages, calls, or links, even if they appear legitimate. Never share 2FA codes.
Scrutinise videos for unnatural movements or overly generous offers, which may indicate deepfakes.
Deny camera access requests from unverified sites and avoid uploading signatures to unknown platforms.
Limit sharing sensitive details online, such as document photos or sensitive work information.

The post Phishing evolves with AI and stealth appeared first on ProtectionWeb.

According to a recent Kaspersky study, the number of cyberattacks experienced by organisations in the last 12-months is reported to have increased by 29% in South Africa. The most ubiquitous threat came from phishing attacks, with 67% of those questioned in South Africa reporting this type of incident. With AI becoming a more prevalent enabler for cybercriminals, over half of the respondents in South Africa (53%) anticipate significant growth in the number of phishing attacks. In this text, Kaspersky examine how AI is used in phishing and why experience alone is sometimes not enough to avoid becoming a victim.

Personalisation through AI

Previously, phishing attacks relied on a generic mass message sent to thousands, hoping some of the recipients would fall for the bait. AI has changed this into scripting highly personalised phishing emails in large numbers. Using publicly available information like that on social media, job boards, and companies’ websites, these AI-powered tools can generate emails tailored to an individual’s role, interests, and communication style. For example, a CFO might receive a fraudulent email that mirrors the tone and formatting of their CEO’s messages, including accurate references to recent company events. This level of customisation makes it exceptionally challenging for employees to distinguish between legitimate and malicious communications. 

Deepfake technology

AI has also introduced deepfakes into the phishing arsenal. These are increasingly being leveraged by cybercriminals to create fake but highly accurate audio and video messages, crafted to reflect the voice and appearance of the executives they seek to impersonate. For example, in one reported case, attackers used a deepfake to impersonate multiple members of staff during a video conference, convincing the employee to transfer approximately $25.6 million. As deepfake technology continues to advance, it is expected that such attacks will become more frequent and harder to detect. 

Bypassing traditional defenses

Cybercriminals can manipulate the script of traditional e-mail filtering systems with the use of AI. By analysing and mimicking legitimate email patterns, AI-generated phishing emails can bypass security software detection. Machine learning algorithms can test and refine phishing campaigns in real time, enhancing their success rates and making them increasingly sophisticated.

Why experience is not enough

Even experienced employees are falling victim to these advanced phishing attacks. The level of realism and personalisation that AI can achieve may override the skepticism that keeps experienced professionals cautious. Moreover, AI-generated attacks often exploit human psychology, such as urgency, fear, or authority, pressuring employees into acting without double-checking the authenticity of the request.

Combatting AI-hyped phishing

To defend against AI-driven phishing attacks, organisations must adopt a proactive and multi-layered approach that emphasises comprehensive cybersecurity. Regular, up-to-date AI-focused cybersecurity awareness training is critical for employees, helping them identify the subtle signs of phishing and other malicious tactics. Kaspersky Automated Security Awareness Platform can help with such training. Alongside this, businesses should implement robust security tools, such as Kaspersky Next and Kaspersky Security for Mail Server, capable of detecting anomalies in emails, such as unusual writing patterns or suspicious metadata.K

A zero-trust security model also plays a vital role in minimising the potential damage of a successful attack. By restricting access to sensitive data and systems, this approach ensures that even if attackers breach one layer of security, they cannot compromise the entire network. Together, these measures create a comprehensive defense strategy, combining advanced technology with vigilant human oversight.

The post The evolution of AI in phishing attacks: Why even the most experienced can fall victim appeared first on ProtectionWeb.