How Switzerland is tackling the AI deepfake menace

AI and social media platforms have accelerated deepfakes Copyright (C) 2026 Ascannio / Shutterstock.

Online deepfakes intend to vilify, mislead or sow confusion and Switzerland is one of many countries seeking to contain the growing problem.

Newer artificial intelligence (AI) tools have drastically increased the scale and scope of deepfakes. The latest AI models make it easy for anyone to create fake images, video or audio content from their own homes without specialist skills.

Recent examples of widespread deepfakes include so-called “nudifier” apps that generate naked pictures of people, including minors. Fraudsters have been conning people out of money using videos of celebrities that are manipulated to show them endorsing fake investments schemes.

Politicians are also increasingly using deepfakes to promote their own images or attack opponents. And there has been growing alarm at foreign states, like Russia, using AI-generated deepfakes to disrupt democratic systems in western countries.

Cybersecurity company DeepStrike says the number of detected cases surged from 500,000 in 2023 to more than eight million last year. The Swiss platform clickandstop.ch, which was set up to record cases of cyberbullying and child sex abuse, revealed a 63% rise in reported cases last year.

“An increasing number of people are also coming forward because photos of them naked have been created and disseminated using artificial intelligence,” said the platform, which is supported by agencies such as Child Protection Switzerland.

Raw statistics alone do not measure the suffering of individual victims, the erosion of public trust in the truth or the threat to democracies. For example, a Tages Anzeiger newspaper investigation uncovered fake sex videos and nude photos of numerous women, including female influencers, being circulated in public Telegram forums.

Such cases have intensified calls to increase the pace and focus of action against the worsening problem. “If we fail to act as a society, it will send a disastrous message: that it’s somehow acceptable to turn real women into virtual sex machines,” stated an opinion piece in the Tages Anzeiger newspaper.

But while there is widespread acknowledgement that the law needs to meet the threat posed by deepfakes to individual integrity and social cohesion, lawmakers in Switzerland and elsewhere, are divided on how to act.

Beefing up legal defences

Swiss officials are divided on the direction, pace and extent of changes to the legal code that are needed to tackle deepfakes. Last year, the Swiss parliament rejected a proposal by Green Party lawmaker Raphaël Mahaim to introduce laws to counter deepfakes.

But in June of this year, the government and the House of Representatives accepted a follow-up motion by Mahaim that specifically targets sexualised deepfakes. This motion followed the scandal of nudifier deepfakes generated by the AI platform Grok, which is integrated into the social media platform X.

Mahaim’s second motion calls for rules to force AI platforms “to conduct risk and damage assessments” of sexualised deepfake content “and to implement protective measures” if needed. The motion also calls for an oversight body that could impose sanctions on platforms that don’t comply.

In earlier parliamentary debates, Swiss government minister Albert Rösti, who heads the communications ministry,  stated that the criminal and civil law codes are sufficient to tackle deepfake abuses. He also pointed out that Switzerland has signed up to the Council of Europe’s AI Convention to defend democracy, the rule of law and human rights against AI abuses. 

As part of this commitment, the justice ministry will offer recommendations on how to tweak some aspects the Swiss legal code. But David Vasella, a partner at the Walder Wyss law firm and founder of the datenrecht.ch platform for data law, does not expect to see a wholesale revision of the legal code with the current government strategy.

“Creating additional laws specifically for deepfakes, which apply on top of existing laws, could create new problems for legal clarity,” Vasella told Swissinfo as he cautioned against complicating the legal code with overlapping laws. “It’s easy to call for additional regulation but hard to do it wisely.”

Instead, Vasella advocates beefing up such bodies as the Federal Data Protection and Information Commissioner, a supervisory authority that protects people from misuse of their data.  
“A new law without adequate enforcement looks good but does nothing,” he said. “Doubling resources for the data protection authority could have a far greater impact. Rather than having new laws we should enforce the existing ones better.”

+ Is AI a risk to democracy?

Holding platforms to account

Some people in Switzerland have been left frustrated by what they see as a vague and slow approach to amending the law. In March, a cross-party group of parliamentarians backed a popular initiative calling for regulations to shield people from digital harm.

If it gathers enough support, the “Internet Initiative” will force a nationwide vote on regulations to protect “fundamental rights and democracy in the digital space”. This includes action against disinformation, sexualised violence and cybercrime.

Technology platforms would be obliged to investigate complaints from the public and to take “necessary countermeasures” against harmful content.

+ Read our series on the ethics of artificial intelligence

The Swiss government has launched its own proposed legislation to hold social media platforms and search engines to account for distributing abusive and false content.

But the draft law typifies the slow rate of progress in this area. First mooted by the Swiss government in 2023, the bill was only introduced last October and has yet to reach parliament. The draft legislation, aimed mainly at large American technology companies, was delayed during tariff negotiations with the US administration last year.

It intends to force the largest digital platforms to formally process complaints of defamation, insults, discrimination and incitement to hatred. Companies would be obliged to set up legal representation in Switzerland and to publicly justify their decisions on removing content or blocking accounts.

The Swiss NGO Algorithm Watch has welcomed the draft legislation as “crucial”. But the pressure group warns that it must be enforced effectively so it “does not simply become a toothless paper tiger”.

Using technology to reduce digital harm

But no matter what politicians decide to do, tech companies also play a role in containing deepfakes. Several tech firms, including Swiss firms, have specialised in detecting deepfakes by tracking trails of digital clues that point to manipulation. But newer AI models are learning to make more believable deepfakes, forcing detectors to constantly upgrade their sleuthing methods. This has resulted in a cat-and-mouse game as both sides try to stay ahead of the curve.

Scientists at the Swiss Federal Institute of Technology ETH Zurich are approaching the problem from a different angle by proving the authenticity of original content with the help of encrypted digital signatures. The institute’s researchers have designed a new type of sensor technology for cameras and other recording devices that imprints a type of watermark onto digital content the moment it is created. This allows people to tell the difference between authentic, original content and any future, manipulated copies.

“Deepfakes are a ‘denial of truth’ attack,” Fernando Cardes, co-creator of the ETH Zurich cryptographic digital signature system, told Swissinfo. “We are providing anchor points of authenticity to help people accurately determine what is real or not.”

The basic principle is not new. The Coalition for Content Provenance and AuthenticityExternal link (C2PA) – a global consortium of technology, media and publishing firms – has been developing different means of authenticating for digital content, including cryptographic signatures, since 2021.

But ETH Zurich researchers believe their provenance system is an improvement on others that imprint signatures onto software platforms. Separating data capture from signature generation gives bad actors the opportunity to drive a wedge between the two with a hardware hack, according to Cardes.

“It’s important to generate signatures onto the same sensor chip that captures the data,” he said. “This makes tampering significantly harder and raises the bar compared to other approaches.”

Having publicly presented their solution, the Swiss researchers now face the task of persuading camera manufacturers and other companies to adopt the technology.

Edited by Gabe Bullard/VdV/ac