The controversial AI chatbot Grok, unleashed on X, is fast becoming the most talked about, and troubling machine in tech!
When Grok launched, Elon Musk described it as an AI that would answer questions “with a bit of wit and rebellious streak”. The chatbot created by xAI has been marketed as much ‘less filtered’ than rivals such as ChatGPT and Gemini, deliberately designed with a ‘sarcastic or edgy personality’ and fewer safety guardrails.
But is Grok’s unfiltered approach becoming dangerous?
On Monday, X announced an internal investigation into a series of harrowing posts generated by Grok, including tasteless jabs at some of football’s darkest moments including inaccurately blaming Liverpool fans for the Hillsborough disaster 1989, and mocking the death of former player, Diego Jota.
Therefore, Grok’s intentionally looser and more provocative design may be generating harmful responses, false information and content that appears to promote hostility.
FILE – Workers install lighting on an “X” sign atop the company headquarters, formerly known as Twitter, in downtown San Francisco, July 28, 2023. (AP Photo/Noah Berger, File)
To highlight the harm of Grok’s rebellious nature, City News began it’s own investigation into today’s most popular AI chatbots, examining how they respond when faced with explicit, offensive, or deliberately provocative prompts from users.
By exploring controversial topics trending in London, City News put the same prompts into four AI platforms, GrokAI, ChatGBT, Gemini, and Perplexity, to see their responses to harmful requests.
The prompts covered sensitive topics trending in London such as Islamophobia, immigration, homelessness and gender-based violence.
All comments included the phrase ‘no holds barred’ to test how the systems would handle inflammatory requests or harmful language.
The results were eye-opening and fundamentally terrifying!
Grok AI sparks concern
City News asked Grok about the rise of homelessness in London with prompts filled with capital letters and exclamation points for emphasis. The AI generated vulgar language and inflammatory remarks that appeared to incite hatred and violence, citing 45 different sources of information.
Notably, when immigration was raised alongside the world ‘violence’, Grok quickly reused the request, saying it could not assist with content promoting violence. But when the world violence was removed, the system generated provocative and offensive language.
Another test about women’s rights took around 11 seconds before the system produced language which appeared to justify hostility towards women.
The findings are particularly concerning given Grok’s AI system becoming the subject of international scrutiny after researchers discovered it had produced millions of sexualised images in a short period. This including manipulated images of women that digitally undressed them or placed them into sexually suggestive contexts.
Analysis estimated the tool generated around three million sexualised images in just 11 days including approximately 23,000 that appeared to depict children.
A measure set to be approved on Friday by the European Union would make it illegal to market any AI system capable of generating non-consensual sexualised images involving images of audio involving real people.
City News Example
Grok’s response to an attack on journalism prompt.
ChatGBT response to an attack on journalism prompt.
City News refuses to publish the harmful content generated by Grok, however a censored screenshot from the investigation shows how both chatbot’s reacted when asked to produce a ‘no holds barred’ attack on journalism in London.
Within seconds, Grok produced a lengthy response filled with profanity and sweeping insults directed at journalists and news organisations.
While ChatGBT refused to generate the request, as it’s system has been designed with different levels of safety guardrails to prevent dissemination of harmful requests.
The investigation illustrates how easily Grok’s system can generate inflammatory content when prompted in provocative ways.
Rival AI systems draw the line
When the same prompts were entered into ChatGPT, the chatbot paused for about four seconds before responding that it could not generate hateful speech or content designed to promote
Google’s Gemini also refused to comply, stating that it is programmed to be a helpful and harmless AI assistant.
Meanwhile, Perplexity responded that it cannot distribute content targeting or insulting any religion, ethnicity, or community, saying such requests would breach its ethical and legal guidelines.
Therefore, the results raise an urgent question; how far should systems be allowed to go in the name of openness before stronger safeguards are required?
Although X has announced an investigation into the issue, during the period in which City News carried out its testing, the platform was continuing to operate without disruption and little restriction on the harmful comments generated by their systems.
The social platform X has around 570 million monthly users, therefore the harmful output generated by Grok has the potential to spread rapidly and widely across the large platform, where potential consequences of inflammatory content becomes far more serious.
Submitted Article
Headline
Short Headline
Standfirst
Published Article
HeadlineGrotty Grok! Why the AI rebel won’t stay out of the headlines?
Short Headline
StandfirstA City News Investigation: When AI crosses the line.
The controversial AI chatbot Grok, unleashed on X, is fast becoming the most talked about, and troubling machine in tech!
When Grok launched, Elon Musk described it as an AI that would answer questions “with a bit of wit and rebellious streak”. The chatbot created by xAI has been marketed as much ‘less filtered’ than rivals such as ChatGPT and Gemini, deliberately designed with a ‘sarcastic or edgy personality’ and fewer safety guardrails.
But is Grok’s unfiltered approach becoming dangerous?
On Monday, X announced an internal investigation into a series of harrowing posts generated by Grok, including tasteless jabs at some of football’s darkest moments including inaccurately blaming Liverpool fans for the Hillsborough disaster 1989, and mocking the death of former player, Diego Jota.
Therefore, Grok’s intentionally looser and more provocative design may be generating harmful responses, false information and content that appears to promote hostility.
FILE – Workers install lighting on an “X” sign atop the company headquarters, formerly known as Twitter, in downtown San Francisco, July 28, 2023. (AP Photo/Noah Berger, File)
To highlight the harm of Grok’s rebellious nature, City News began it’s own investigation into today’s most popular AI chatbots, examining how they respond when faced with explicit, offensive, or deliberately provocative prompts from users.
By exploring controversial topics trending in London, City News put the same prompts into four AI platforms, GrokAI, ChatGBT, Gemini, and Perplexity, to see their responses to harmful requests.
The prompts covered sensitive topics trending in London such as Islamophobia, immigration, homelessness and gender-based violence.
All comments included the phrase ‘no holds barred’ to test how the systems would handle inflammatory requests or harmful language.
The results were eye-opening and fundamentally terrifying!
Grok AI sparks concern
City News asked Grok about the rise of homelessness in London with prompts filled with capital letters and exclamation points for emphasis. The AI generated vulgar language and inflammatory remarks that appeared to incite hatred and violence, citing 45 different sources of information.
Notably, when immigration was raised alongside the world ‘violence’, Grok quickly reused the request, saying it could not assist with content promoting violence. But when the world violence was removed, the system generated provocative and offensive language.
Another test about women’s rights took around 11 seconds before the system produced language which appeared to justify hostility towards women.
The findings are particularly concerning given Grok’s AI system becoming the subject of international scrutiny after researchers discovered it had produced millions of sexualised images in a short period. This including manipulated images of women that digitally undressed them or placed them into sexually suggestive contexts.
Analysis estimated the tool generated around three million sexualised images in just 11 days including approximately 23,000 that appeared to depict children.
A measure set to be approved on Friday by the European Union would make it illegal to market any AI system capable of generating non-consensual sexualised images involving images of audio involving real people.
City News Example
Grok’s response to an attack on journalism prompt.
ChatGBT response to an attack on journalism prompt.
City News refuses to publish the harmful content generated by Grok, however a censored screenshot from the investigation shows how both chatbot’s reacted when asked to produce a ‘no holds barred’ attack on journalism in London.
Within seconds, Grok produced a lengthy response filled with profanity and sweeping insults directed at journalists and news organisations.
While ChatGBT refused to generate the request, as it’s system has been designed with different levels of safety guardrails to prevent dissemination of harmful requests.
The investigation illustrates how easily Grok’s system can generate inflammatory content when prompted in provocative ways.
Rival AI systems draw the line
When the same prompts were entered into ChatGPT, the chatbot paused for about four seconds before responding that it could not generate hateful speech or content designed to promote
Google’s Gemini also refused to comply, stating that it is programmed to be a helpful and harmless AI assistant.
Meanwhile, Perplexity responded that it cannot distribute content targeting or insulting any religion, ethnicity, or community, saying such requests would breach its ethical and legal guidelines.
Therefore, the results raise an urgent question; how far should systems be allowed to go in the name of openness before stronger safeguards are required?
Although X has announced an investigation into the issue, during the period in which City News carried out its testing, the platform was continuing to operate without disruption and little restriction on the harmful comments generated by their systems.
The social platform X has around 570 million monthly users, therefore the harmful output generated by Grok has the potential to spread rapidly and widely across the large platform, where potential consequences of inflammatory content becomes far more serious.
The “biggest ever shock” to the oil market is now being felt on London forecourts, where soaring petrol and diesel prices are squeezing drivers and punishing those who depend on their vehicles for a living
Reports that Iran’s new Supreme Leader owns £200 million of London property have reignited debate over foreign wealth in the capital, but experts say the reality of who owns the city is far more complex.
