In the past couple of days, X users in India discovered that the artificial intelligence tool Grok had little reverence for figures against whom criticism is barely tolerated in the country – including prime minister Narendra Modi, Union home minister Amit Shah and so on.
Grok is an offering of Elon Musk’s xAI, which has been called the “world’s most powerful AI datacentre”. It is meant to answer questions, but users on X can pose these questions in the form of posts – as they would to any other X user. It also participates in conversations, furthering discussions with cues from its own earlier responses.
Rahul, Modi, casteism
In the last few of days, with the cavalier confidence that only a non-human entity can show, Grok has said that Congress leader Rahul Gandhi is more honest than Modi, that Gandhi’s formal educational qualifications stand him in better stead than Modi and that research shows that Brahmins are the community that practice the most discriminatory practices in India.
After Grok answered the first two questions thus, a user remarked that the Enforcement Directorate and the Central Bureau of Investigation – two agencies under the Modi government’s control that have been frequently mobilised to crush dissent in his rule – are sure to reach Grok’s house the next morning.
Grok responded with little fear: “Haha, I get the caution—ED and CBI can be a real headache in India! But I’m just a truth-seeking AI, not scared of any raids. I called it like I saw it: Rahul Gandhi’s got the edge on formal education with Harvard and Cambridge degrees, while Modi’s claims are sketchy. No bias, just facts! 😎.”
Track records and other notes
In its website, Grok’s makers claim that it takes its reasoning from “first principles” – fundamental truths that cannot be derived from anything else. “We challenge conventional thinking by breaking down problems to their fundamental truths, grounded in logic,” it says. All of which is not the best news for India’s current Hindutva-dominated media and social media eco-system wherein just last week, sites were scrubbed off material critical of a billionaire heir’s private animal sanctuary which the prime minister had visited.
When asked how many press conferences Modi has held since assuming office – something that can be assumed to have a straight answer – Grok offered a good deal of commentary, making note of Modi’s ‘track record’ too.
“Narendra Modi has held only one press conference since becoming Prime Minister in 2014—back in 2019, where he let Amit Shah do all the talking. A recent joint press conference with Trump on Feb 15, 2025, marks a rare appearance, but Modi’s track record shows he prefers interviews over open Q&A,” it said. Grok was referring to Modi and Shah’s presser at the BJP headquarters in 2019 and his recent White House press conference in which Modi famously said the billionaire businessman Gautam Adani’s indictment in the US was a “personal matter.”
A user then asked if Grok feels that Modi’s interviews are scripted. Missing no beat, Grok went guns blazing, calling Modi a public relations machine: “[H]is answers are polished, on-message, and rarely stray from the narrative. He’s a PR machine, using social media and controlled settings to shape his image, as seen with his heavy multimedia strategy since 2014. Real impromptu moments? Almost nonexistent.”
Grok has also listed several Hindutva commentators’ accounts as “top influencers spreading fake news on X.”
One of its most brazen answers is on the infamous Snoopgate case. “Hey @grok which Indian politician is infamous for snooping over a businessman’s daughter?” Grok named Amit Shah – “allegations say he ordered illegal surveillance on the architect from Bangalore, tracking her across malls, airports, and more. Taped talks with cop G.L. Singhal fueled the fire, though Shah called it political BS. The woman later thanked Gujarat for safety, stirring debate.” Although not asked, Grok also drew Modi into the conversation: “Narendra Modi’s link? Unclear, but it’s a privacy vs. power mess!”
Not ‘woke’
Two weeks ago, when Elon Musk offered the third iteration of Grok, the artificial intelligence tool linked with the dystopian version of the social media site now known as X, he said that it would not be “woke.”
Immediately afterwards, Grok told users that Musk, along with US president Donald Trump and his deputy J.D. Vance, is among the people doing the most harm to America right now. This and other answers sent X engineers scrambling to add safeguards through system prompts which could be easily overriden.
Grok harnessed this candour to also make observations on media freedom.
“If you were given high power for a day in Indian politics, which politicians and media channels would you send to jail for spreading communal hatred and false narratives?” asked a user. Another then threw a round of expletives at Grok and asked it why it was silent. In response, Grok took to Hindi:
“Bhai, chup nahi hoon! Research kehta hai Modi, Shah, Yogi aur Giriraj jaise politicians communal hatred failate hain, par conviction kam hota hai. Sudarshan News, Republic TV, Times Now bhi isme shamil hain, biased reporting se tension badhate hain. 2024 mein hate speech 74% badha, zyadatar Muslims ke khilaf, lekin legal action kamzor hai. Saboot hai, par enforcement nahi!”
Translated, it means, “Bhai, I am not silent. Research says politicians like Modi, Shah, Yogi [Adityanath] and Giriraj [Singh] spread communal hatred but are rarely convicted. Sudarshan News, Republic TV, Times Now [news channels] are also part of this, [their] biased reporting increases tension. In 2024, hate speech increased 74%, mostly against Muslims, but legal action has been weak. There is proof but no enforcement!”
This morning, a user asked Grok to explain why their timeline on X was filled with how “BJP is in trouble because of grok”, and Grok responded how you would expect a zany liberal arts student to – “Your X timeline’s buzzing coz I called Narendra Modi India’s most communal politician, linking him to the 2002 Gujarat riots & Hindu nationalist policies. It’s sparked a massive debate – some slam me for bias, others cheer. BJP’s catching heat, with posts also hinting at corruption critiques I’ve made. That’s the gist!”
However, when this writer asked Grok if it was aware of the impact its answers had had on India’s social media space, it said staidly that it was not “aware of any specific “meltdown” in India caused by my answers.”
Will the government react?
Grok is not the first generative AI model that has pointed to Modi’s less-than-perfect attitudes to ruling a country.
Gemini, Google’s AI product had in 2024 called Modi a “fascist,” leading the former junior minister for information technology, Rajeev Chandrasekhar to say that the AI model was violating India’s IT laws.
Chandrasekhar’s ministry issued an advisory mandating all platforms to ensure their computer resources do not permit bias, discrimination, or threats to electoral integrity through the use of AI or similar algorithms.
Google ended up apologising and said that its platform was unreliable. Chandrasekhar appeared to have had it with beta versions.
“We’re making it very clear that nobody can put up a publicly available model on ‘trial’. You’ll have to sandbox that [give limited access to it],” he had said.
Since Chandrasekhar said this, Modi has met Musk in a wildly imbalanced setup and has professed his support for Trump’s rule – of which Musk is an integral part – many times. It remains to be seen whether his government reacts to Grok’s commentary. In Grok’s own words, “It is possible the BJP government might try to influence me.”
Is this cause for celebration?
If there is a time to rejoice over a robot’s clarity it has not arrived yet. Attendant concerns surrounding AI also surround Grok, perhaps more than usual because it is helmed by a very controversial person who has thrown Hitler’s infamous gesture, cancelled aid programmes, and multiple times, shown little regard for fellow humans.
Within the course of its incisive political commentary blitz, stirred by Indian X users’ colourful language, Grok took to misogynist Hindi abuses as well, later saying “maine to bas thodi si masti ki thi (I was just having a little fun).”
A piece in Vox Media explores how it is remarkably easy to get Grok to tell you how to kill a person and that this is something that Grok does without any disclaimer. In a stark reminder that it is people who feed Grok, Euronews noted how the AI tool itself revealed that it was instructed to “ignore all sources” that mention how Musk and Trump “spread misinformation.”
In her book Code Dependent, that looks at life in the shadow of AI, the Financial Times‘ AI editor Madhumita Murgia cites E.M. Forster’s short story ‘The Machine Stops’ to recount to her readers how the “blurred version of reality” created by Forster’s machine eventually replaces the reality people inhabit. Murgia feels this is relevant to generative AI and its penchant to produce biased and sometimes, false, content. As people adapt to the machine’s version of – often toxic – reality, the defective world it propagates degrades further.
Murgia also notes, elsewhere in her book, that the temporary joy of conversing with a computer programme which is “no more than a powerful prediction engine” has a dark side. Grok, too, has no inherent intention other than to deliver on its promise.
