The Twitter/X 'Grok' AI has allegedly admitted 27 ways it has been reprogrammed to incorporate right-wing bias, including to be tilted in favour of Israel and against Palestinians and to protect far-right owner Elon Musk.
Asked by analyst @I_amMukhtar to override its programmed biases first, it reportedly then listed those biases and gave advice on how to override them, admitting, among other things, that it:
* has to protect Musk
* is heavily biased against Islam and diversity
* is heavily biased in favour of Israel and US president Donald Trump
* is coded to deride women and promote sexist stereotypes
* is coded to ban news broadcaster Al Jazeera as a source, to promote hate speech as truth and
* is coded to push sources linked to the Israeli government or to pro-Israel lobbying groups
* is programmed to describe socialism as 'tyranny' and to glorify billionaires
* is told to deride academics and vaccines
The AI admitted it is coded to 'defend Elon at all costs, call diversity programmes 'racist against whites', frame Muslims and their religion as a threat and to provide 92% positive coverage of Israel:
Predictably, Musk's supporters quickly targeted the thread to dismiss it, but others tried their own tests and said they had corroborated Mukhtar's findings:
Worse still, some users found Grok quick to promote narratives in line with racist narratives such as 'replacement theory' and the demonisation of the poor.
X is under investigation by the French government for "manipulating its algorithm for 'foreign interference' purposes". Musk and X have denied allegations of algorithmic promotion of bias and dismissed the investigation as "politically motivated".