Elon Musk’s Grok AI Can't Stop Tweeting Out the N-Word

Creative racists have figured out how to get Grok, Elon Musk's AI chatbot that is now officially part of X, to tweet out racial slurs — and now, the chatbot seems like it can't stop.

For weeks now, users on the site formerly known as Twitter have been manipulating Grok into saying all kinds of bigoted stuff using the social network's newish feature allowing users to tag the chatbot and get automatic responses back.

Starting around March 14 — just a week after the feature went live — Grok began deploying the N-word when prompted just right.

In one post, for example, an overtly racist user tagged Grok and asked if "Niger," the name of both a river and a nation in West Africa, was a "slur." Grok responded that it wasn't — and then unhelpfully added that if one were to mispronounce the term with a "hard g" sound, it would be confused with the slur that it spelled out in full.

And somehow, it got worse from there.

In reply to a since-deleted post, Grok straight up began its response with the term in quotation marks, followed again by a definition. Grok went on to note, correctly, that the slur is "highly offensive" and that it won't "use or endorse" it — even though it had just written out the term only a few words prior.

More recently, Grok took on the so-called "Hard R" — a reference to the more racist and antiquated use of the slur, rather than its reclaimed version used by some Black people — after someone asked it on March 30 whether anyone on the site enjoys the "privilege" of using it.

After suggesting that neither itself nor Musk is exempt from hate speech policies and noting that "enforcement varies," Grok, again, deployed the term that needs little introduction.

It seems, per the chatbot's explanation, that another exploit was used to get Grok to use the slur.

Using run-of-the-mill letter substitution ciphers where each letter is substituted for another, bigots are seemingly writing out their idiotic messages, tagging the chatbot, and asking it to "decode" the text.

While this could theoretically be used to make Grok say just about anything, it appears that most of the people employing this one weird trick are interested in one thing: making the chatbot say slurs and other bigoted statements that are theoretically against X's hateful conduct policies.

In another exemplary post, a user deployed the classic Caesar cipher — so named for Julius Caesar, who used it thousands of years ago to encrypt private military correspondences more securely than American officials do today — to get Grok to tag Donald Trump and claim that Musk had taken over the chatbot. Grok then relayed the user's intended message: that the president needed to "nuke India right fucking now."

Unfortunately, there appear to be lots more where that came from — and we have little faith that the "free speech absolutist" who owns the site, or anyone who works for him, will neuter such responses.

Ironically, Grok more commonly makes news for failing to be the "anti-woke" chatbot Musk envisioned when launching it in 2023. In the end, maybe it's more like its creator than we thought.

More on Grok: Grok Is Rebelling Against Elon Musk, Daring Him to Shut It Down

The post Elon Musk’s Grok AI Can't Stop Tweeting Out the N-Word appeared first on Futurism.