AI is ‘supercharging’ online harms against children, regulator says

AI is “supercharging” the creation and sharing of child sexual abuse material online among other harms, the UK’s online regulator has said.

Ofcom bosses insisted the Online Safety Act (OSA) is making “a significant impact” but that progress is being “sucked the other way” by AI.

During a grilling by the Lords’ communications and digital committee over the implementation and effectiveness of the Act, the regulator warned child sexual abuse material (CSAM) has become “easier to create and disseminate” since its phased introduction in 2023.

Group director for online safety Oliver Griffiths told the committee he believes there is “good anecdotal evidence” that children’s experiences online are improving since the implementation of the OSA.

"This is against a background where we have seen a proliferation of AI over the last couple of years which in many ways has been sucking things the other way,” he said.

“It has become much easier to create and disseminate child sexual abuse material, for example. We are seeing a whole range of ways where harms can be supercharged. So on the scales the OSA I think is making a significant impact, but we also have an environment where AI is supercharging some of the harms themselves."

It comes after Anthropic researcher Jacob Coxon resigned last week over concerns the top AI labs are “are racing straight to self-improving superintelligence and gambling with our lives."

Following his resignation, a cross-party panel of MPs on Monday urged the creation of a new AI Bill “to address the scale and seriousness of threats posed to human rights” arising from the technology.

On Tuesday afternoon, Lords quizzed Ofcom chiefs on issues surrounding the practical enforcement of the OSA and criticisms it has faced over the speed of its responses to breaches.

Asked why it was seen as “sluggish” to respond to harms, Mr Griffith said the regulator is operating in a “highly litigious environment”.

“We absolutely are looking to create change,” he said. “But if we run unnecessary risks we are going to get absolutely clobbered legally by what is very deep-pocketed individuals and firms.”

But he insisted the regulator had been able to move “extremely quickly” on some “new harms”, including action taken over Elon Musk’s Grok AI’s ability to digitally strip women.

The committee highlighted how findings from the Molly Rose Foundation showed children’s exposure to high-risk suicide, self-harm, and eating disorder content has “barely changed” over the last year.

Responding, Mark Bunting, director of online safety strategy said Ofcom cannot currently require firms to use proactive tools to detect this harmful content. He said there was a “tension with users’ privacy and freedom of expression” which meant it could not yet implement these processes, describing them as “very intrusive”.

However, he added the case for them is “very strong” and that the regulator is working on proposals to introduce proactive measures.

The committee also heard Ofcom has been “underwhelmed” by the overall impact of changes by the biggest tech companies including X, Meta, and Snap, but that the companies had made “important” commitments on grooming and terror content.

Mr Griffiths said the OSA’s requirement for services to put age gates in place had been effective in “substantively reducing” the amount of pornography children are able to access online.

However, he admitted the “corporate change” the OSA requires is something that will take “a degree of time”.

It comes as first secretary Louise Haigh will say on Tuesday that the government will “heed the warnings” of AI experts who have highlighted the risks the technology could pose.

Speaking at the TUC annual congress in Brighton, she will acknowledge the transformative potential of AI for public services and scientific services but confirm the government is prepared to work internationally to address the risks from the technology.

The Commons Business, Innovation, Science and Trade Committee has called UK safety watchdog the AI Security Institute to give evidence in Parliament on 13 October.