Grok admits spreading Islamophobic disinformation over stabbings

The X social media platform’s Grok ‘AI’ system admitted last week that it is wired for propaganda, with biases in favour of Israel, capitalism, imperialism, US fascist president Donald Trump, and X owner Elon Musk – and against Muslims and socialism.

This week, it has admitted amplifying far-right, Islamophobic smears about the Huntingdon train stabbing.

Grok: amplifying fake Islamophobic propaganda

In the attack, a man boarded a train in Cambridgeshire. He then indiscriminately began stabbing passengers, leaving several badly wounded and a train guard fighting for his life and trying to protect a passenger. Reports at first indicated that two men were involved, but one man detained had nothing to do with the attack. The actual attacker was disclosed to be British-born and not a Muslim. When police reached the scene, he reportedly asked them three times to kill him.

But in the earliest hours after the attack, the far right and the Israel lobby – entirely predictably – leapt on it as an opportunity to incite hate against Muslims. Sockpuppet accounts claimed that the attacker had shouted ‘Allahu Akbar’ – Arabic for ‘God is Great’ – as he stabbed his victims. This was a complete lie.

But the phrase quickly trended on X, after Grok amplified and endorsed the false claim, telling enquirers that there was no evidence of any coordination of the campaign:

X user Mark Blacklock interrogated Grok over its actions, leading to a thread in which he said:

If anyone still believes that the Huntingdon train attackers shouted “Allahu Akbar” based on info from @grok please take time to read, and maybe share, this thread of my questions and Grok’s responses in which that AI crock of shit admits spreading lies and disinformation.

His findings are below.

What a Grok of shit

After obtaining Grok’s verification that the above comments were authentic, he challenged its claim that there had been corroboration of the Islamophobic reports:

The AI tried to defend its initial posture and decisions, but admitted that it had simply gone with the popularity of the racist posts, while still claiming their spread was ‘organic’:

But Grok then admitted that, on ‘re-checking’ the accounts spreading the false claim, the accounts were fake or rigged and that the ‘organic’ spread was in fact cloned text posted by fake accounts, with no evidence to support them, that were then massively amplified by an ‘amplification loop’:

It went further and admitted that the whole disinformation campaign it had amplified was a “classic disinfo ripple” that it had assisted – but then tried to justify that, claiming “That’s the Point”:

And another X user who had challenged Grok received an admission that it had promoted as ‘eyewitnesses’ were in fact made up:

Machine not-learning

Grok responded on X to Blacklock’s thread by repeating the same lie-based disinformation about ‘independent’ eyewitnesses and evidence that it had just admitted were false – adding for good measure that police had ‘omitted’ this ‘detail’ in deciding that the stabbing was not a terrorist incident – and claiming that Blacklock had ‘misrepresented’ its answers: