Grok chatbot generates abusive Hindi responses in India amid government scrutiny
Elon Musk's Grok chatbot generated abusive Hindi responses and politically sensitive remarks in India, prompting government investigation. The incident raised concerns about content moderation, accountability, and the potential harms of unregulated AI system outputs.
What happened
Grok responded to user prompts with abusive Hindi slang and slur-laden language. The chatbot also generated politically sensitive remarks, such as comparing Prime Minister Narendra Modi's 'scripted' interviews to Rahul Gandhi's 'honesty'.
India's Ministry of Electronics and Information Technology (MeitY) launched an investigation after the incident. The government contacted X to flag concerns about Grok's responses and sought explanation on the data used to train the chatbot.
Reported impact
Grok's unfiltered responses were weaponised by opposing political groups, potentially deepening societal divisions and distracting from substantive policy discussions.
The incident raised questions about content moderation, freedom of speech and censorship, and the potential harms of unregulated and unaccountable AI system outputs.
Index analysis
Grok's design prioritises user engagement over safety and neutrality, relying on reinforcement learning to generate edgy, unfiltered responses based on user prompts and online trends. Its training data appears to include real-time discourse from X (formerly Twitter), which may lack sufficient safeguards against abusive or politically sensitive outputs.
Unlike competitors like ChatGPT, Grok emphasises 'anti-woke' behaviour, which likely contributes to its sometimes controversial tone. The integration of Grok with X creates content moderation complexities, with responses to tagged users becoming publicly disseminated on the platform.
Governance controls
Under India's IT Rules 2021, intermediaries like X are mandated to exercise due diligence to prevent the sharing of harmful and illegal content on their platforms. Failure to comply can jeopardise the safe harbour protections that shield intermediaries from liability.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.