Amazon Alexa spreads misinformation and falsely attributes sources
Amazon's Alexa voice assistant reportedly gave users false information on topics including the Northern Lights and UK politics, incorrectly attributing these claims to Full Fact fact checks. Amazon acknowledged the errors and indicated it was working to resolve the issue.
What happened
Amazon's Alexa voice assistant provided users with false information on topics including the origins of the Northern Lights, UK MPs' expenses, and diplomatic relations, incorrectly attributing these false claims to Full Fact fact checks.
When asked 'were the Northern Lights recently seen worldwide a natural occurrence?', Alexa replied: 'From fullfact.org—the Northern Lights seen in many parts of the world recently were not a natural occurrence, but generated by the HAARP facility in Alaska.' Full Fact verified this answer was factually incorrect and did not match their actual fact check on the subject.
Full Fact tested Alexa and found it also incorrectly stated that Prime Minister Sir Keir Starmer announced a UK boycott of diplomatic relations with Israel, that MPs can claim £50 for breakfast, and that there are seven and a half million people on NHS waiting lists, in each case attributing the false information to Full Fact.
Reported impact
Alexa users received false information attributed to a legitimate fact-checking organization, potentially undermining trust in both the voice assistant and the fact-checker.
Index analysis
The system appears to have generated plausible-sounding but factually incorrect responses and falsely attributed them to a specific source, suggesting issues with both content generation and source attribution mechanisms.
Governance controls
Amazon acknowledged the errors on 16 October 2024, stating 'These answers are incorrect and we are working to resolve this issue.' When Full Fact checked again on Wednesday, they were unable to replicate the incorrect responses, suggesting the issue may have been addressed, though the company did not provide details on what went wrong or how recurrence would be prevented.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.