Politics

Copy article

New law to tackle Gen AI child abuse images at source as rate of production doubles

ended 13. November 2025

World-leading legislation sees UK Government work with AI industry and child protection organisations to stamp out criminality. 

The UK government announced today legislation enabling authorised bodies to test AI models for child safety vulnerabilities before harmful content can be generated. This marks a significant shift from reactive removal to proactive prevention.

The Internet Watch Foundation reports of abusive AI-generated material more than doubled from 199 cases in 2024 to 426 in 2025. 

Current law prevented safety testing because possessing such material, even for protective research, was illegal. This created an impossible bind: models couldn't be stress-tested until harm had already occurred.

The new legislation designates AI developers and child protection organisations as authorised testers, allowing them to identify and close vulnerabilities at source. Category A content (the most severe) now represents 56% of all illegal AI material, up from 41% last year.

We'd like your views:

  • Does pre-release testing actually stop determined offenders, or will they simply circumvent safeguards through jailbroken or unregulated models?
  • How do we adequately protect researchers and moderators conducting this testing from psychological harm and legal vulnerability?
  • Should tech companies face financial penalties if their models are later exploited despite passing initial safety tests?
  • How can regulators possibly keep pace when new AI models launch every month, each requiring thorough vetting?

1 responses from the Newspage community

Copy all

Star Quote
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
While this legislation is welcome, pre-release testing won't stop determined offenders. They'll just build their own jailbroken systems or use unregulated open-source models. What it does do is make exploitation harder and more traceable for casual misuse, which is where most harm starts. The real test is whether we're building detection systems that flag when safeguards are being systematically bypassed, because deterrence only works if circumvention actually gets caught.

Cybersecurity firm Group-IB, which collaborates with international law enforcement agencies like INTERPOL and Europol, reports this isn't theoretical. Their analysis of the dark web already shows that criminals are actively building and using 'uncensored' models to bypass commercial safeguards for all sorts of illegal activities.