Securing safe online spaces for minors - Have the G7 effectively regulated generative AI?

Criminology student, Eloise Taylor, reflects on discussions about AI regulations at the 52nd G7 Summit in Evian, France.

Leaders of the G7 Summit pose for an official photograph
Off

Summary 

The proliferation of AI-generated Child Sexual Abuse Material (CSAM) is a direct consequence of inconsistent and ineffective generative AI regulations. This analysis piece discusses the G7’s reserved approach to AI regulations at the 52nd G7 Summit in Evian, France. It also outlines previous beneficial G7 human-centric commitments but argues there is still a long way to go before regulations successfully secure a safe online space for minors.  

The Evolution of AI and CSAM 

Since 2023, generative AI models no longer require large software to access, they are now utilised on mundane devices for everyday activities. Unfortunately, actors have exploited this easy access to create CSAM.  

This is a growing international issue. INTERPOL, ECPAT and UNICEF identified 1.2 million children across 11 countries who had their images exploited via generative AI models in 2025, with children continuously re-victimised through the production of this AI generated material. This re-use makes law enforcement action difficult to improve due to difficulties determining which children are in immediate danger or if it is re-victimisation.  

Unfortunately, there is a lack of consistent international AI regulation to ensure all technological companies are building safe models. For example, in America, some states have strict regulations, whereas other states leave it to the companies to ensure pre-market risk assessments are conducted. The UN Secretary General highlighted how this leads to children becoming test subjects for unregulated models and has allowed individuals too much AI power. Rightly, the CEO of the Internet Watch Foundation is now calling for all Governments to set a “non-negotiable standard in AI development”.  

Where’s the G7? 

The 2023 G7 Hiroshima Summit was the first significant step towards global governance of generative AI with the approval of the Hiroshima AI process (HAIP). This introduced 11 voluntary codes of conduct for generative AI developers and included crucial principles on ‘safety-by-design’ approaches. This includes tech companies conducting external independent testing of AI models prior to release and then from the test results implementing appropriate mitigation measures that could prevent misuse of models. 

During Italian Prime Minister Georgia Meloni’s 2024 presidency of the G7 the HAIP reporting framework was further developed in cooperation with the OECD. This reporting framework monitors the implementation of the HAIP codes, which helps with accountability of tech companies and shows how companies are operationalising the guidelines, thus assisting in identifying best practices. At the 2025 summit, the HAIP was further expanded to include SMEs, with the aim of raising awareness of the codes to smaller companies, therefore helping them to build more trustworthy models. Yet, there is still a clear reluctance amongst the G7 to incorporate voluntary guidelines into mandatory regulations.

There are benefits to this flexibility, including rapid advancement of AI law enforcement tools that could tackle CSAM. However, as with the disaster of Elon Musk’s Grok AI model, which produced 23,000 CSAM in 11 days, there are also high risks associated with a lack of regulations and this is a global issue that must extend beyond the membership of the G7. Whilst the G7 HAIP friend group has involved other countries in AI regulation, there is still a globally fragmented response, resulting in AI offences being displaced to another country, with less regulation. 

France 2026: Simply a continuation of the voluntary AI Approach?

As with previous summits, AI regulation was rightfully a central topic. Under the theme of ‘Ensuring a safe, fast and efficient deployment of artificial intelligence’, Macron focused on the multitude of benefits AI innovation can bring to countries but appropriately balanced this with the clear risks to children. It was the first time a G7 Summit addressed AI regulation in relation to online child safety, showing a positive global recognition that the two issues are closely interlinked.  

It was also encouraging to see Macron championing coordination across countries as a crucial way to tackle AI-generated CSAM. To ensure this cooperation transpired at the summit, Macron secured Trump’s attendance, a key stakeholder in AI developments, until the end of the summit, with promises of a Versailles dinner on the final evening. He also ensured that Trump could not discard any discussions by not having a final communique and by phrasing AI regulation for the benefit of children rather than a standalone topic. 

Further, Macron invited Prime Minister Modi, and other countries, into the AI debates, including in his closing summit remarks that digital ministers are building a platform between democracies affected by AI. This will be used for coordination on common standards in AI and for the sharing of best practices in cybersecurity. He also set the standard for future summits through inviting AI tech giants, including Open AI and Antrophic CEOs, to have a seat at the table. This is critical to ensuring accountability of technological companies and to understand their views on online child protection. 

However, Trump still managed to find a way to overshadow this promising format by banning the use of Anthropic’s top AI models by non-Americans directly before the summit. This highlighted to the G7, that during Trump's tenure in power, access to powerful and US AI models is no longer a foregone conclusion. None of the G7 were going to waste the opportunity to address this whilst the American administration and the CEO of Anthropic were in the same room, which unfortunately moved talks away from child safety to AI sovereignty, further watering down the focus on CSAM.

While some progress was made on safe digital space for minors, the declaration on this and the G7 Common Set of Principles for secure digital space for minors, did not advance AI regulatory standards with the main focus remaining on voluntary principles only. The G7 did correctly identify that we need to educate children and show parents how to navigate the online world. However, this cannot be the main solution to online child abuse. It wrongfully places the onus on the victims, rather than the technological companies, who should be creating safe spaces that proactively prevent online CSAM in the first place.  

Further whilst voluntary regulations have worked for large tech companies, such as Google, smaller companies are finding loopholes because they do not have the resources to implement the guidelines. Therefore, introducing international mandatory regulations, similar to the EU AI Act (2024) requirements, is not simply desirable, but a necessity. 

This may be difficult for the G7 to accept given the millions invested in AI developments, with Keir Starmer recently announcing a £1.1 billion AI investment plan, yet effective mandatory regulations would decrease the costs of harms created by CSAM, with France estimating the annual cost of sexual violence agaisnt children at €9.7 billion, and the lifetime economic costs of child victims of sexual violence and abuse in the UK estimated to be £148 billion. However, the G7 can still offer a forum in which leaders could come to a cohesive position that cuts-across powerful states on developing regulatory frameworks that would have a global impact on CSAM.

Conclusion 

Overall, the Evian Summit has improved the possibility of international regulatory cohesion in generative AI developments in the future, with the French presidency attempting to set the agenda more firmly onto this issue which may offer some reassurance for children, parents, social care services and law enforcement. Unfortunately, the G7 will be hosted by the US next year putting the promising progress since 2023 at risk of following a Trump focused deregulation agenda that threatens to undermine the progress made by the G7 in the last three years.

Further Reading