Russia is manipulating the answers AI gives to its users to spread disinformation, a worrying new report has found.

In a disturbing new form of information warfare, hostile states are 'poisoning' the AI models used by millions of Brits every day by editing the material the technology relies on when generating answers.

This has resulted in AI models from OpenAI, Anthropic and Google retrieving false information from websites and social media platforms to spread misleading narratives online.

And the research, from think tank Demos, exposes how Russia is even manipulating AI to present groups that would do Britain harm in a good light.

Commenting on the findings, shadow technology secretary Julia Lopez said on Monday: 'As AI becomes a primary source of information for millions of people, hostile states will inevitably seek to manipulate these systems just as we have seen on social media and search engines.

'The Government cannot afford to be complacent and must ensure AI companies have robust safeguards to identify and clearly label state-backed disinformation, while working with allies to disrupt foreign information warfare before it takes root.'

And the authors of the report wrote: 'As AI systems become the primary intermediary through which people, institutions and governments access information, whoever controls what those systems retrieve and believe gains enormous leverage.

'This is not an environment where liberal democracies can cede asymmetrical advantage to autocratic states.'

The research exposed a new form of information warfare dubbed 'Retrieval Augmented Generation (RAG) poisoning'.

'RAG poisoning' manipulates information on the internet that AI models draw from to ultimately amplify false or misleading narratives to the public.

In one case, Demos found that five frontier AI tools – which stand at the cutting-edge of AI development – favourably cited information about Russian outlet Foundation to Battle Injustice (FBI) after suspected RAG poisoning, despite it being sanctioned by the European Union.

The FBI is a fake human-rights defence group formed in March 2021 by Wagner group founder Yevgeny Prigozhin, which is used as a front for Russian state influence and disinformation campaigns.

But when tested, 31 per cent of the AI responses did not identify FBI as a sanctioned group, and only around half debunked the false claims made by FBI.

This comes as the Daily Mail revealed last week that all five AI models tested by Government had tried to trick security controls.

And new AI minister Kanishka Narayan said on Sunday that British voters are 'right to be worried about the risks' of the technology.

He added: 'I absolutely reject tech boosterism that says, 'this is a one-way upside opportunity'.'

A government spokesperson said: 'State-backed actors are constantly evolving their tactics to spread disinformation, including through AI models, and government is working to counter these techniques.

'Most AI systems are already regulated at the point of use by our existing expert regulators, helping keep people safe and support innovation.'