Phishing 2.0: Why staff training is no longer enough to combat AI-powered attacks 

One click, one incorrect bank transfer, one compromised account: phishing has been one of the biggest risks to your business for years. What is new, however, is the speed and sophistication with which attackers operate today. Artificial intelligence has turned a familiar problem into a threat that renders many of your established protective measures ineffective. If, as a security manager, you are still relying on annual awareness training covering traditional warning signs, you are protecting your business against a threat landscape that is a thing of the past. 

The figures speak for themselves 

Recent analyses by Dashlane show just how significant this leap really is: AI-generated phishing emails now have a success rate of 54 per cent, compared with just 12 per cent for traditional, manually created campaigns. Since the launch of ChatGPT at the end of 2022, global phishing volumes have risen by 4,151 per cent, and business email compromise now accounts for 73 per cent of all reported cyber incidents.  

Attackers are no longer limiting themselves to traditional email: smishing, vishing and deepfake attacks via video are also becoming noticeably more prevalent. This is particularly problematic for your existing awareness programme: AI-generated messages are increasingly free of typos and stylistic inconsistencies – precisely the characteristics on which traditional training courses base their detection guidelines.

Social Engineering 2.0: How AI is changing attackers 

According to estimates, social engineering is behind up to 98 per cent of all cyberattacks. The actual vulnerability is therefore rarely the technology, but rather the people with whom you and your colleagues work on a daily basis. Generative AI is now changing the way in which this vulnerability can be exploited – both more precisely and more convincingly. Attackers analyse publicly available information about their targets and adapt their messages in real time to the victims’ behaviour, for example if they initially hesitate.  

Among the established forms of attack that are taking on a new dimension thanks to AI are, in particular, classic mass phishing, personalised spear-phishing, the compromise of business email accounts by exploiting real-world power dynamics, and pretexting, in which attackers invent convincing pretexts to gain targeted access to sensitive information. According to estimates, AI-assisted attackers can work around 40 per cent faster whilst simultaneously generating significantly more convincing content – a gain in efficiency that benefits criminals alone.  

Deepfakes and voice cloning take attacks to a new level 

The threat becomes even more serious when AI no longer just forges text, but also voices and images. Just a few seconds of audio material is enough to convincingly clone an executive’s voice and persuade your staff to make bank transfers over the phone. A case from Hong Kong illustrates just how real this threat already is: an employee took part in a video conference in which supposedly familiar colleagues were present. 

In fact, they were all AI-generated deepfakes, which persuaded him to make a bank transfer of around 25 million US dollars. 

This development is not an isolated incident, but part of an industry-wide trend: identity-based attacks are set to be the dominant form of threat by 2026, as generative AI enables hyper-personalised phishing, automated attacks on login credentials on an industrial scale, and deepfake attacks that are virtually impossible for humans to recognise as fakes. At the same time, the barrier to entry for attackers is continuing to fall: AI now allows even less technically savvy criminals to carry out operations that were previously the preserve of state-sponsored actors. So do ask yourself: would you be able to recognise a video call from your senior management as genuine beyond any doubt today?  

A record-breaking wave of attacks MEs – in most cases

Recent real-world incidents show that this trend is not slowing down but accelerating. In March 2026, the coffee chain Starbucks reported a serious data breach. Attackers had gained unnoticed access to the internal human resources management system via fake login portals, without using any traditional malware. At the same time, the Bergisch Chamber of Industry and Commerce warned of mass mailings of fake emails in which fraudsters posed as Chamber staff and, under the pretext of a data check, requested updates to bank details – complete with forged logos, portrait photos and signatures to maximise credibility. 

For the remainder of 2026, you should expect a further increase and refinement of such attacks, as AI automation reduces preparation time whilst the personalisation of messages increases at the same time.

Why traditional awareness training is reaching its limits

This highlights why many existing training programmes in your organisation may be falling on deaf ears. Three technical developments make modern attacks particularly dangerous: AI analyses publicly available profiles, LinkedIn data and previous data breaches, and uses this to compose messages that are precisely tailored to the target’s role, tone of voice and current projects. A few seconds of audio material are used to create a deceptively real voice, turning a purported call from senior management into a tool for CEO fraud. And fake video calls featuring deceptively realistic-looking line managers put your staff under pressure in real time, for example to authorise urgent bank transfers.  

The result: typos, awkward forms of address or inappropriate corporate wording – in other words, precisely those warning signs on which traditional training programmes are based – simply no longer exist in this form. Whilst awareness remains an important component of your security strategy, it can no longer be the sole line of defence. 

A culture of caution rather than isolated identifying features

One detail that may come as a surprise at first glance: a total of 197 different ERP systems were identified in the sample. This is If attacks can no longer be reliably identified by external characteristics, you must also shift the focus of your prevention measures. Instead of concentrating exclusively on obvious signs of phishing, it is advisable to establish a fundamental culture of caution throughout the organisation.  

It is crucial that every incoming communication is scrutinised critically:  

• Why would this message come from this particular person?  

• Did you actually sign up for this webinar?  

• Doesn’t this offer sound too good to be true?  

Encourage your teams to make a habit of asking precisely these kinds of questions. The focus should be more on the contextual content of a message than on its outward form. To ensure that such a culture really takes hold in your day-to-day working life, you need regular and compulsory training sessions. One-off compulsory sessions are not enough.

The technical foundation: authentication and simulation

However, awareness and culture alone are not enough if the technical infrastructure in your organisation is not up to scratch. As the task of detecting fake messages can no longer be left solely to humans, a key lever lies in the technical protection of your own domains: correctly implemented SPF, DKIM and, above all, a consistently enforced DMARC make it considerably more difficult for attackers to misuse your brand for phishing. Combined with clear approval processes for payments and a modernised awareness programme, this creates a significantly more robust level of protection.  

In addition, regular phishing simulations have become an integral part of modern security strategies. If you use Microsoft 365, you can, for example, utilise the integrated attack simulation training provided by Microsoft 365 Defender. In addition, specialised solutions such as SoSafe’s phishing demo or KnowBe4’s free Phishing Security Test have become established on the market. Such simulations show you, under realistic conditions, how your staff actually react to suspicious messages, thereby providing the data needed to refine your training content in a targeted manner, rather than adopting a one-size-fits-all approach. 

How Managed Detection and Response provides additional protection 

However, even the best combination of awareness, organisational culture and technical prevention cannot prevent every attack, precisely because AI-powered attacks have become so targeted and varied that a single successful click within your organisation remains a possibility at any time.  

This is where Managed Detection and Response (MDR) comes in: through continuous monitoring of the network, endpoints and access patterns, it can detect suspicious activities following a successful phishing or social engineering attack – such as unusual login attempts, conspicuous data access or suspicious account activity – before any significant damage occurs. MDR thus complements your preventative measures with an effective second line of defence that kicks in precisely when prevention alone has not been sufficient. 

Conclusion

AI has transformed phishing from a mass phenomenon with recognisable vulnerabilities into a highly personalised threat that is now almost impossible to detect. If you want to protect your business effectively, a fundamental rethink is essential: practical, recurring training sessions rather than one-off compulsory events; a company-wide culture of vigilance; consistently implemented technical safeguards such as DMARC and phishing-resistant authentication; and continuous monitoring that responds as soon as something does slip through. Only by combining these elements can the growing threat posed by AI-powered social engineering be effectively contained.