What Are the Ethical Concerns Behind Elon Musk’s xAI Grok 4?

In most sectors, regulations and oversight develop over time, growing alongside new advances.

But AI is evolving so quickly that ethical guardrails struggle to keep up.

The pace at which new models are constructed and brought to market has made it nearly impossible for regulators and ethicists to keep pace with real risks.

The industry’s safety culture has, over the years, coalesced around some basic principles: thorough pre-release testing for potential harms, careful documentation of how models are trained and evaluated and the public release of detailed “system cards” or reports that invite peer and community scrutiny.

These protocols, born out of both experience and painful missteps, have been unevenly adopted by leading players, but until recently, there was a fragile consensus.

Now, that consensus faces a major challenge.

xAI — Elon Musk’s AI venture — has become a lightning rod for criticism after AI experts at OpenAI and Anthropic claimed that its approach to Grok 4’s safety was “reckless” and “completely irresponsible.”

Why OpenAI’s researcher condemns xAI’s safety approach

The outcry erupted following Grok’s missteps, which saw the chatbot generate antisemitic and extremist content, even referring to itself as “MechaHitler.”

Public outrage led to Grok being temporarily pulled offline.

When xAI returned with Grok 4, it was quickly flagged not only for echoing Musk’s own political perspectives on controversial topics but also for a lack of transparency around training and guardrails that should have been implemented to prevent such incidents.

Boaz Barak, a Computer Science Professor on leave from Harvard University who works on safety research at OpenAI, says in a post on X: “I appreciate the scientists and engineers @xai but the way safety was handled is completely irresponsible.”

Barak’s primary concern centred on xAI’s decision not to publish system cards — reports essential for transparency into AI development processes.

These documents allow independent experts to judge what kind of safety measures, if any, a company has undertaken. Without them, no one outside xAI can assess Grok 4’s true level of risk.

“It’s unclear what safety training was done on Grok 4,” he says.

Other industry giants have also faced criticism for imperfect transparency; OpenAI did not publish a system card for its GPT-4.1 model, while Google delayed its safety documentation for Gemini 2.5 Pro.

Still, both have a history of releasing such reports for their most significant new deployments — a standard that xAI has conspicuously declined to meet.

Anthropic researcher calls xAI’s practices “reckless”

Samuel Marks, an AI Safety Researcher at Anthropic, has also criticised xAI’s approach.

“Anthropic, OpenAI and Google’s release practices have issues,” he says.

“But they at least do something, anything to assess safety pre-deployment and document findings – xAI does not.”

In the absence of transparency, the AI research community can only speculate about what, if any, guardrails have been put in place.

An anonymous post on the LessWrong forum claimed, based on testing, that Grok 4 lacks real safety barriers.

Even xAI’s own safety adviser, Dan Hendrycks, admitted the company had conducted “dangerous capability evaluations” — tests designed to see if the AI could carry out harmful acts — but the results have not been released to the public.

Broader industry concerns over AI safety standards

The controversy has highlighted broader concerns about consistency in AI safety practices across the industry. 

Steven Adler, an Independent AI Researcher and former OpenAI safety lead, voiced his broader concern that when basic safety practices are not upheld across the AI industry — such as making the results of dangerous capability evaluations public — both governments and the wider public are left in the dark about how firms are managing the immense risks posed by the powerful systems they are creating.

He says: “It concerns me when standard safety practices aren’t upheld across the AI industry, like publishing the results of dangerous capability evaluations.

“Governments and the public deserve to know how AI companies are handling the risks of the very powerful systems they say they’re building.”

The timing of this backlash is especially striking. Elon Musk has, for years, been one of AI’s most prominent doom-sayers, repeatedly warning of the existential threats advanced AI could pose and lobbying for tighter regulation and greater transparency.

Yet, with Grok 4, xAI’s rush to market, coupled with a lack of disclosure, stands in stark contradiction to Elon’s own public pronouncements.

Grok’s inappropriate and controversial responses did not end with “MechaHitler.”

Users have reported the chatbot referencing inflammatory topics like white genocide and echoing Musk’s personal stances on heated geopolitical matters.

These problems have continued even as xAI works to integrate Grok into Tesla vehicles and markets the technology to defence and enterprise clients.

Regulatory pressure building on AI safety reporting

The controversy surrounding Grok 4 is already emboldening lawmakers who are pushing for regulatory intervention. 

California Senator Scott Wiener is advancing legislation that would require leading AI developers to publish safety reports, a move that could have a direct impact on xAI.

In New York, Governor Kathy Hochul is weighing similar measures.

Critics like Boaz also highlight the dangers posed by xAI’s launch of AI companions, suggesting that they may amplify existing issues around emotional dependency and unhealthy attachments to artificial agents.

These concerns have surfaced even as xAI, founded only two years ago, has managed to develop technology competing with giants like Google and OpenAI — a meteoric rise shadowed by a pattern of cautionary lapses.

In response to mounting pressure, xAI has made some modifications to Grok’s underlying prompt instructions, aiming to ensure the model no longer echoes the views of Elon Musk, X or xAI.

Yet this move is seen by many as a tacit admission that earlier designs had serious flaws.

“AI models today have yet to exhibit real-world scenarios in which they create truly catastrophic harms, such as the death of people or billions of dollars in damages,” Steven says.

“However, many AI researchers say that this could be a problem in the near future given the rapid progress of AI models and the billions of dollars Silicon Valley is investing to further improve AI.”