63% of UK AI companies don’t disclose whether they train models on user data

The information on UK AI companies comes from the recently released Cybernews AI Trustworthiness Ranking 2026, which assesses 500 leading AI companies across four pillars: security, data privacy, transparency, and public perception.

The security pillar examines whether a company has a dedicated security or trust page, a bug bounty program, and an ISO/IEC 27001 or SOC 2 certification.

The data privacy pillar analyzes companies’ privacy policies to determine whether they clearly state what user data they collect, whether user data is used for AI model training, third-party data-sharing practices, and data retention practices.

The transparency part looks at whether companies disclose their legal entity and physical address, and have a working, non-automated contact channel.

And finally, the public perception score is based on company reviews from Trustpilot and G2.

Out of the 500 AI companies analyzed, 19 were based in the UK, and the average Trustworthiness score of these UK AI companies was 64, which is one point below the global average.

Information on model training and data retention is lacking

Almost half (47%) of the analyzed United Kingdom AI companies do not mention anything in their privacy policies about training their AI models on user data, and 16% are vague about it.

In addition, 58% of the top United Kingdom AI companies do not say precisely how long they retain user data: 11% don't mention anything about retention or deletion, while 47% only mention it vaguely, without any precise retention periods.

Primarily because of these issues, the average privacy score is 67 out of 100 across the 19 UK AI companies we investigated–three points below the global average.

“People use AI tools for everything these days, and they give those tools access to their most sensitive data. That’s why it’s so important for users to clearly understand what is done with their data. People shouldn’t have to interpret vague language or struggle to understand how their data is being used,” says Voldemaras Kadys, member of the AI Trustworthiness Ranking Advisory Board and Head of Security and Platform Engineering at Mediatech, the digital publishing house behind Cybernews

Low security scores are concerning

The lowest scores among UK AI companies are by far in the security pillar, which means many companies lack a dedicated security or trust page, a bug bounty program, and security certifications.

UK-based AI companies’ security scores averaged out to 29–also three points below the global average.

In our methodology, security has a modest 20% weight because we can only assess a company’s security from the outside, so our view is limited, yet it’s still concerning that companies handling some of the most sensitive conversations do not have these security elements in place.

Organizational transparency was the strongest-performing pillar, with an average score of 90 out of 100.

The public perception score, based on user reviews from Trustpilot and G2, averaged out to 74 across the 19 UK AI companies, slightly above the global average of 71.

Best and worst-scoring industries

There’s a huge difference between the top and worst-performing categories in our analysis.

Companies in the image generation & editing market had an average Trustworthiness score of 91, whilst businesses in the AI detection & anti-detection category had an average score of just 49.

Synthesia and Recraft received the highest scores

Synthesia (AI video platform for business) and Recraft (text-to-image generation) received the highest Trustworthiness scores among UK AI companies, both scoring 91 out of 100.

They were followed by ElevenLabs, which scored 86, VEED.IO with 81, and Paintit.ai with 80.

Explore the full AI Trustworthiness Ranking: https://cybernews.com/ai-knowledge-base/ai-trustworthiness-ranking-2026/