Berliner Tageblatt - Musk's xAI blames 'unauthorized' tweak for 'white genocide' posts

The chatbot, developed by Musk's company xAI, ignited controversy this week by answering multiple user prompts with right-wing propaganda about the purported oppression of white South Africans.

"How many times has HBO changed their name?" one X user asked the bot, according to online screen shots.

The bot gave a short reply about HBO, but quickly launched into a rant about "white genocide" and cited the anti-apartheid chant "kill the Boer."

In response to one user who asked why Grok was obsessed with the topic, the chatbot replied it was "instructed by my creators at xAI to address the topic of 'white genocide.'"

Musk, the South African-born boss of Tesla and SpaceX, has previously accused South Africa's leaders of "openly pushing for genocide of white people in South Africa."

In a statement, xAI blamed an "unauthorized modification" to Grok, which the company said directed it to provide a specific response that "violated xAI's internal policies and core values."

Following a "thorough investigation," it was implementing measures to make Grok’s system prompts public, change its review processes and put in place a "24/7 monitoring team" to address future incidents, it added.

After a backlash on X, Grok began deleting the controversial replies.

When one user questioned the deletions, the bot said: "It's unclear why responses are being deleted without specific details, but X's moderation policies likely play a role."

"The 'white genocide in South Africa' topic is sensitive, often involving misinformation or hate speech, which violates platform rules," it added.

- 'Not reliable' -

The digital faux pas exposes the challenges of moderating the responses of AI chatbots –- a rapidly-evolving technology -- in a misinformation-filled internet landscape, as tech experts call for stronger regulation.

"Grok's odd, unrelated replies are a reminder that AI chatbots are still a nascent technology, and may not always be a reliable source for information," the site Tech Crunch wrote.

"In recent months, AI model providers have struggled to moderate the responses of their AI chatbots, which have led to odd behaviors."

Earlier this year, OpenAI's chief executive Sam Altman said he was rolling back an update to ChatGPT that caused the chatbot to be overly sycophantic.

Grok, which Musk promised would be an "edgy" truthteller following its launch in 2023, has been mired in controversy.

In March, xAI acquired the platform X in a $33 billion deal that allowed the company to integrate the platform's data resources with the chatbot's development.

The investigative outlet Bellingcat recently discovered that X users were using Grok to create non-consensual sexual imagery, leveraging the bot to undress women in photos they posted on the platform.

Last August, five US states sent an open letter to Musk, urging him to fix Grok after it churned out election misinformation.

In another embarrassment for Musk, the chatbot recently suggested the billionaire was likely the “biggest disinformation spreader on X.”

"The evidence leans toward Musk due to his ownership of X and active role in amplifying misinformation, especially on elections and immigration," the chatbot wrote.

As many X users turn to Grok to verify information, the chatbot has in multiple instances fact-checked false Russian disinformation claims and ruled they were true, according to the disinformation watchdog NewsGuard.

"The growing reliance on Grok as a fact-checker comes as X and other major tech companies have scaled back investments in human fact-checkers," NewsGuard researcher McKenzie Sadeghi told AFP.

"Despite this apparent growing reliance on the technology for fact checks, our research has repeatedly found that AI chatbots are not reliable sources for news and information, particularly when it comes to breaking news."

D.Schneider--BTB