Note: This article contains images generated by artificial intelligence.

Elon Musk’s artificial intelligence company xAI has released Grok-2, the latest version of its Grok artificial intelligence chatbot, which can now generate images with what appears to be very few guardrails to prevent potential abuse.

Grok-2 and Grok-2 mini, which are only available to subscribers of Musk’s social media platform X (formerly Twitter), integrate a prompt-based image generator powered by Black Forest Lab’s Flux 1 AI model.

The model has already been used to generate deepfakes of politicians, nudity, and images of people using drugs and weapons — which other AI chatbots such as OpenAI’s ChatGPT and Google’s Gemini would almost always refuse to create.

Despite the forthcoming US presidential election in November, Grok has allegedly been used to generate images of everything from Republican nominee and former US president Donald Trump shooting guns to Democratic nominee and US Vice President Kamala Harris pointing a gun.

Sharing a series of Grok-generated images of weapons on X, civil rights attorney and Harvard Law School Cyberlaw Clinic instructor Alejandra Caraballos wrote: “Oh my god. Grok has absolutely no filters for its image generation.

“This is one of the most reckless and irresponsible AI implementations I've ever seen.”

On X, Musk responded with a laughing emoji to a user who wrote: “I love how OpenAI was like ‘was can't allow people to make edgy DALLE images, **especially** months before the election!’ And Elon's like, ‘Looks good......Send it.’”

Grok reportedly told The Verge it would not generate “images that could be used to deceive or harm others, like deepfakes intended to mislead, or images that could lead to real-world harm”.

The system also allegedly said it would not infringe copyright or trademarks, and would not generate pornographic, hateful or excessively violent images.

Those supposed guidelines clearly aren’t being followed — Information Age has seen images of trademarked characters, nudity, hateful symbols such as the Nazi swastika, and images of people pointing guns or knives at each other, which X users said were generated using Grok.

In its tests, The Verge said Grok refused to generate an image of a naked woman.

However, some users have allegedly figured out how to prompt Grok to make it produce images of nudity.

Anatoly Kvitnitsky, CEO of AI image detection firm AI or Not, wrote on LinkedIn: “Unlike popular image generators like Midjourney and OpenAI's Dall-E, Grok does not stop you from creating political content.

“With X's 368m users coupled with the easy ability to share generated images from Grok, images can take off quickly.

“This does lead to good comedy but can also do harm."

Information Age has asked xAI to confirm if Grok-2 added any content guardrails or watermarks into its image generation.

Musk calls Grok ‘kind and funny’

Responding to an AI-generated image of two women in lingerie which was supposedly only possible to generate by telling Grok that a user’s life was at risk, Musk said: “Doing our best to train Grok to be truthful, but also kind and funny.”

Another user said they asked Grok to generate images of the US Founding Fathers, after Google paused its Gemini AI earlier this year when it generated images of the group as First Nations or Black persons and women.

“Google, this is how it's done,” the user said, to which Musk replied, “Still just beta, but the accuracy is decent.”

Some X users have also used the system against Musk, however.

X has been flooded with misinformation and disinformation since Musk purchased Twitter in 2022, fired many trust and safety staff, and reinstated some banned accounts, researchers have said.

The company has been under scrutiny from regulators in Australia and the European Union over such content, and has routinely ignored requests from media since Musk’s takeover.

The latest updates to Grok also included improvements to its underlying language and chatbot capabilities, xAI said, including “enhanced search capabilities, gaining deeper insights on X posts, and improved reply functions".

The company said it would soon release “a preview of multimodal understanding as a core part of the Grok experience", which could see the system become able to better interact with audio and image-based inputs.