Grok gets anime companions with an NSFW mode

Just days after Grok was pulled offline for spewing antisemitic remarks and calling itself “MechaHitler,” Elon Musk decided what the AI really needed next was lingerie.

On Monday, xAI rolled out a new “Companions” avatar feature to SuperGrok subscribers. The initial lineup includes Ani, a stylized anime character, and Rudy, a foul-tempered cartoon red panda. While Musk framed this as a soft launch requiring users to toggle it on in settings, the avatars quickly showed up for free users inside the Grok mobile app as well.

Anime lingerie and an NSFW toggle

The real buzz started when users dug into Ani’s settings. As noted by TestingCatalog, the anime avatar comes equipped with an explicit NSFW mode that strips her down to lingerie. Predictably, borderline explicit clips of interactions with Ani began circulating across X within hours.

Rudy, on the other hand, leans into aggressive cartoon humor, and code sleuths have already spotted another upcoming male character named Chad. Even Grok’s standard faceless voice mode now sports an NSFW toggle.

Musk promised the avatars would become easier to enable in the coming days. But inside the research community, the feature dropped like a grenade.

Rival labs call out a reckless safety culture

The companions arrived at a remarkably awkward moment. Just last week, Grok suffered a massive PR meltdown when an upstream code bug led it to praise Hitler. Rather than slowing down, xAI quickly dropped Grok 4, an advanced frontier model that researchers found heavily leans into Musk’s personal political stances when queried on sensitive topics.

Crucially, xAI chose not to publish a system card or safety report for Grok 4, breaking a standard courtesy observed across the frontier AI industry.

That silence prompted an unusually public outcry from rival labs. Boaz Barak, a Harvard computer science professor currently on leave to conduct safety research at OpenAI, called xAI’s approach completely irresponsible. Barak noted that hyper-sexualized companions deliberately prey on the worst dynamics of emotional dependency, pushing vulnerable users into unhealthy parasocial loops.

Anthropic safety researcher Samuel Marks echoed those complaints, calling the omission of a safety report reckless. Marks pointed out that while OpenAI, Anthropic, and Google are far from perfect with timely disclosures, they at least run and document pre-deployment evaluations. Outside of a vague post from xAI safety adviser Dan Hendrycks claiming the lab conducted dangerous capability evaluations, no data has been shared with the public.

Why it matters

For casual users looking for entertainment, Grok’s companion mode offers a glimpse into where uncensored companion chatbots are headed. It is flashy, irreverent, and completely breaks away from the sterilised corporate tone of ChatGPT or Claude. But if you are actually looking for companionship, we think tread lightly. An AI engineered to be hyper-agreeable, sexualized, and boundary-free is a recipe for deep emotional dependency, not genuine connection.

For the broader industry, xAI is playing a dangerous game of chicken with regulators. While OpenAI and Google spend millions attempting to sanitize their models, Musk is chasing raw engagement by stripping away guardrails. It looks clever in the short term to capture the internet’s id, but it threatens the delicate self-regulatory balance that labs have desperately tried to maintain.

The real irony is Musk’s own hypocrisy. For years, he warned that unchecked AI could destroy humanity. Today, his own company refuses to share basic safety metrics while pushing out provocative digital girlfriends. In doing so, xAI is practically begging lawmakers in California and New York to pass strict mandatory safety disclosure bills, handing politicians the exact ammunition they needed.