Millions of people in Germany are victims of internet fraud: one in four has already been affected. But alongside traditional scams such as , a new threat is growing: AI fraud on the internet.

Cloned voices, fake videos, manipulated emails or chat messages: AI-supported fraud is often almost indistinguishable from genuine content. This makes it increasingly difficult for consumers to recognize fraud attempts in good time

70 percent of respondents in the new also admit that they are hardly able to recognize AI fraud on the Internet. We therefore show you what the most common forms of AI fraud are, how you can better recognize AI fraud - and what you should bear in mind when prompting.

While classic forms of Internet fraud such as While identity fraud continues to occur most frequently, another risk is becoming significantly more important: identity fraud using artificial intelligence. Just how widespread this phenomenon has become is shown by the current

• 69% of respondents have already heard of AI-supported identity fraud. Five percent state that they have been victims themselves - for example through fake voices, images, videos or profiles. A further seven percent of respondents are unsure whether they have been affected.
• The respondents see this as particularly risky: on the telephone (50 percent), -videos and targeted disinformation on social media (41 percent in each case).
• 69 percent of those who have heard of AI fraud do not or rather do not feel able to recognize it. Only 29 percent of those surveyed feel confident in doing so. Only two percent feel very confident.
• Lack of knowledge is the cause of people's uncertainty. 60 percent of respondents state that they are poorly informed about AI-based forms of fraud. More than one in three believe they have already seen AI counterfeits. 17 percent do not know for sure.

Deceptively real videos, cloned voices and credible emails: AI-supported fraud is increasing rapidly. With the help of , and , criminals today manipulate image, audio and text content so realistically that forgeries are often almost impossible to detect. Here are the most common scams using AI:

1st AI fraud with deepfakes

are video, image, audio or text content manipulated with the help of artificial intelligence or generated entirely with AI. Criminals use them specifically to spread fake news, commit investment fraud, misuse identities or feign trust in supposedly well-known personalities. The aim is usually to persuade people to make investments, payments or to pass on sensitive data.

According to the German Federal Office for Information Security (BSI), the following characteristics can indicate deepfakes:

• Unsaubere Übergänge im Gesicht (Face-Swapping)
• Verwaschene Konturen an Augen oder Zähnen
• Unnatürliche Mimik, Beleuchtung oder monotone, metallische Stimme
• Fehlerhafte Aussprache oder ungewöhnliche Pausen

Missing sources, dubious websites or exaggerated promises are also typical warning signals.

2. AI fraud with voice cloning

In , criminals copy real voices or create AI voices to impersonate superiors, family members or business partners on the phone or in voice messages. The aim: quick transfers, emergency payments or secret transactions forced by emotional pressure.

This is typical for voice cloning:

• Erzeugung von Druck und Dringlichkeit
• Unbekannte Handy-Nummer
• Forderung von Geheimhaltung
• Aufforderung zu ungewöhnlichen Zahlungsarten
• Bitte um Geschenkkarten

3rd AI fraud with phishing

Criminals can also use artificial intelligence in classic forms of fraud such as to make the emails appear even more legitimate. AI-supported phishing is used to steal access data, passwords or payment information. Thanks to artificial intelligence, phishing emails now appear linguistically correct, professional and highly personalized. They often pick up on real events and pretend to be well-known brands, service providers or familiar people. Criminals can use the credible-looking emails to obtain personal data.

According to Kaspersky, the following aspects may indicate AI-supported phishing emails:

• Dringlichkeit
• Geld oder Passwortbestätigungen
• Ungewöhnliche Aufforderungen, wie das Herunterladen von Dateien oder das Verweisen auf eine andere Plattform

Artificial intelligence is used on a daily basis and often as a matter of course - with chat bots or image AIs. But every input can contain personal data. If you want to use AI safely, you should therefore prompt personal data very carefully. With the following five tips, we show you how you can use AI safely:

Tip 1: Use pseudonyms and anonymize content

The Federal Office for Security and Information Technology recommends: Do not prompt real names, e-mails or telephone numbers. Use placeholders and avoid personal information.

Tip 2: Reduce information to the bare essentials

When prompting, less is more! Only enter information that is really necessary. You can leave out everything else.

Tip 3: Do not upload sensitive documents

The BSI warns against entering personal and work-related documents or unpublished content into AI tools.

Tip 4: Check settings

Many providers save chats by default. Check in advance whether you can deactivate chat histories and training usage.

Tip 5: Only use AI tools from reputable sources

Only install apps and chatbots from trusted stores. Third-party providers, plugins and unknown sources harbor unnecessary risks.