Grok inspired Misinformation reveals a premature dependence on “fact Checking AI”
X (Twitter) AI technology bot “Grok” managed to insert itself into the middle of American politics this week for all the wrong reasons after it provided incorrect responses that formed the basis for many high-profile posts by US politicians, resulting in misinformation being shared widely on the X platform and beyond.
What happened?
Over the last week a number of protests – dubbed “No Kings” – were held across US cities in protest of many of the Trump Administration’s policies and actions during Trump’s second term as US president. Many local and national news outlets covered the protests, including MSNBC which included various aerial shots of the protests, as seen below.
The above image, which was part of a video segment, aired on MSNBC during the protests. The segment was widely shared on social media by MSNBC, its affiliates as well as many influencers and politicians.
But then unfounded claims began to surface online claiming that this aerial shot that featured on MSNBC during the “No Kings” protests was actually taken during the Women’s March from January 2017 – also in Boston – and that MSNBC was re-purposing old video in order to lie to the public and mislead viewers as to how many protestors showed up for the 2025 “No Kings” protests.
Those unfounded claims that MSNBC was re-using old video were amplified thanks to X’s very own ‘Grok’. As X users in their thousands turned to the AI bot to find the truth about the source of the MSNBC video, the AI bot spent much of last week telling users that yes, the aerial shot used by MSNBC in 2025 was actually from 2017, and MSNBC was misleading its viewers.
This then led to a number of high profile politicians and influencer accounts accusing MSNBC of lying, including Texas Senator Ted Cruz (since deleted his post – no correction issued at time of print).
The problem however, is that it was Grok that was wrong. There were no “multiple fact-checks” that the video was from 2017. It really was from October 2025, and the “community note” Grok referenced was simply a proposed Community Note that had not passed any verification checks, nor did it offer any evidence about the video being taken in 2017.
Two days after giving incorrect responses, Grok now does indeed recognise that the footage was taken in 2025, not 2017.
Sponsored Content. Continued below…
While AI bots are progressing at rapid speed, their ability to analyse video footage is still inconsistent, and this latest snafu also reveals that Grok may be too dependent on X’s Community Notes feature. Fact checking these types of claims isn’t usually difficult for human fact checkers, however. The video footage used by MSNBC didn’t exist online before October 2025, as reverse image tools reveal, which is usually a solid indicator that the footage is new. You can also confirm these claims by actually checking video footage of the 2017 Women’s March which, while similar in appearance because it takes place at the same location, is still distinctively different (differing amount of verdure on the trees, for example). And again, you can check the footage of local affiliates and organisations to confirm that yes, the footage MSNBC used was from 2025, not 2017.
And the broader point here is that while AI can be great for many different things – as we can attest to first-hand – it’s influence in the fact checking domain is problematic. AI isn’t foolproof, yet its determinations are already being treated like Gospel for many social media users, when the technology certainly hasn’t done enough to merit that type of treatment, if indeed any single technology can ever be deserving of such blind trust.
TL:DR – Grok and other AI chatbots are a great tool and can certainly help with fact checking – but like with any other single entity, don’t blindly rely on them. They make mistakes. Double and triple verify.
Continued below...
Thanks for reading, we hope this article helped, but before you leave us for greener pastures, please help us out.
We're hoping to be totally ad-free by 2025 - after all, no one likes online adverts, and all they do is get in the way and slow everything down. But of course we still have fees and costs to pay, so please, please consider becoming a Facebook supporter! It costs only 0.99p (~$1.30) a month (you can stop at any time) and ensures we can still keep posting Cybersecurity themed content to help keep our communities safe and scam-free. You can subscribe here
Remember, we're active on social media - so follow us on Facebook, Bluesky, Instagram and X