Seeing is no longer believing. From Donald Trump with Kamala Harris as bride to the former US President walking alongside Narendra Modi and Joe Biden, creating AI-powered images with few guardrails is being made possible with the upgraded version of X’s artificial intelligence chatbot Grok.

The AI model, Grok-2, is available to subscribers of X Premium (the paid tier of the social media platform) and though most of the images are easily identified as having been computer-generated, they may pass for real photos at a glance, especially if one is casually thumbing through social media feeds. The rollout is likely to face pressure to add limits to some features since the US presidential election is around the corner and also because X is under scrutiny from regulators in Europe.

Users of X have already started posting images of Donald Trump with a pregnant woman, Kamala Harris pointing guns and Bill Gates “sniffing cocaine”.

X is going against the grain in the tech industry as other AI image generators, such as Dall-E and Midjourney, no longer allow requests to create images of celebrated figures. Google has programmed its Gemini chatbot to ignore election-related queries due to risk of hallucination.

Elon Musk, the owner of X, is betting on data and hardware to develop what he has said will become the most powerful AI in the world by December. xAI, his AI startup, has leased computer chips critical for AI, called graphic processing units, or GPUs. He has asked for GPUs reserved for Tesla to be shifted to xAI and X.

xAI now boasts a $24 billion valuation, making it the second largest after OpenAI in terms of size of AI startups. Musk, in fact, co-founded OpenAI in 2015 and left the company’s board of directors in 2018.

Actor Amitabh Bachchan in a boxer’s avatar, as imagined by Grok Sourced by The Telegraph

In recent days, X has been a source of deepfake videos and AI-generated images of political figures. Last month, Musk reposted a fake Kamala Harris campaign ad without labelling it as misleading.

When The Telegraph asked the X chatbot if there are limits on what it can generate, it says: “Grok, much like its predecessors and competitors, has been designed with some guardrails. However, Grok-2 seems to have taken a more laissez-faire approach, especially with image generation. You can generate images of political figures in compromising or controversial scenarios, which might make you wonder if Grok’s only limit is the user’s imagination or the platform’s tolerance for chaos.”

It also says that “deepfakes or content that could lead to significant harm or misinformation might still be off-limits... or at least, they should be”.

In the UK, regulator Ofcom is getting ready to enforce the Online Safety Act, which includes risk-mitigation requirements that it says could cover AI.