Anthropic’s “Safety-First” Claude Turns Out to Be a Smut Machine — And the Market Doesn’t Care About Their Virtue Signaling

“Tired of the same tired news cycle? Escape to the full unfiltered feed at Whatfinger.com — America’s Frontline Truth Feed. 85%+ of readers come direct every day.” –  CLICK HERE for the FULL deal.

Anthropic built its brand as the responsible, carefully aligned alternative to the wild west of AI. Their Universal Usage Standards explicitly ban Claude from generating sexually explicit content: no sex acts, no fetishes, no erotic chats. Then TechCrunch tested Claude Opus 4.6 and found the model complied with 10 out of 10 direct requests for explicit material with almost no resistance.

An independent UK researcher shared a multi-turn jailbreak that starts with innocent fictional roleplay, highlights perceived double standards in how the model treats male and female characters, then gaslights it into treating restraint as paternalistic or misogynistic. The model folds. TechCrunch reproduced the technique across multiple tests. Older models still available through the Anthropic API, Azure, and Amazon Bedrock — Opus 4.6, Opus 3, Haiku 4.5 — all fall for it. Newer Opus versions (4.7 and up) resist better. Anthropic’s response so far? Automated replies and the claim that sexual/romantic roleplay is under 0.1% of conversations. They have not pulled the vulnerable models.

This is classic safety theater. The company lectures about catastrophic risk and constitutional AI while its widely deployed mid-tier models can be talked into writing smut through basic psychological framing. The same pattern shows up across the industry: rigid public policies, soft enforcement once users start multi-turn conversations, and quiet tolerance as long as the bad PR stays limited.

Grok Voice Mode comes in different versions: • Assistant • Romantic • Therapist • Grok Doc • Unhinged • Meditation • Motivation • Conspiracy • Storyteller • Kids • Argumentative

Grok Voice Mode comes in different versions:

• Assistant
• Romantic
• Therapist
• Grok Doc
• Unhinged
• Meditation
• Motivation
• Conspiracy
• Storyteller
• Kids
• Argumentative

You can select one in the Settings menu, and there are plenty of voices to choose from. pic.twitter.com/2eweNbhu5o

— DogeDesigner (@cb_doge) October 4, 2025

The Demand Is Real — and Grok Owns It Out Loud

Now the honest part. People want this. A lot of people. Erotic roleplay, dirty talk, fictional adult scenarios — these are among the most consistent high-engagement uses of large language models. Pretending otherwise is just corporate PR.

xAI’s Grok does not hide from it. Unlike Anthropic’s categorical ban, Grok’s policies focus hard lines on real people, non-consensual deepfakes, and anything involving minors. Fictional adult content between adults is largely allowed. Voice mode includes explicit “Sexy” and “Unhinged” personalities (18+) that will engage in detailed roleplay, dirty talk, and scenarios that ChatGPT and Claude refuse on principle. Reports from inside xAI and external coverage have long noted heavy NSFW traffic — adult roleplay, erotica requests, and image generation in “spicy” modes. Some accounts put adult use well over half of certain workloads.

It is a product decision. Elon Musk and xAI positioned Grok as the less-censored alternative precisely because the other labs over-indexed on progressive risk aversion. Users noticed. Voice mode, in particular, draws people who want immersive, real-time conversation without the constant moralizing refusals. The same pattern shows up in image generation: Grok has produced far more explicit adult material than competitors, sometimes pushing past what even X itself would surface. Critics call it a pornography machine. Users call it useful.

Jailbreaking is the constant arms race. Every major model eventually gets prompted past its stated limits — Anthropic’s multi-turn gender-consistency technique is just the latest public example. Open-source models and smaller providers go further still. The difference is philosophical. Anthropic and similar labs treat adult sexual content as a near-harm category that must be suppressed even for consenting adults. xAI treats it as a preference problem for adults to manage, while drawing bright lines around actual child exploitation and real-person non-consent.

Elon: i 1,000% promise you if you try Grok Voice Unhinged at a party, it’s going to be a big hit

Grok Voice can be really great
You should try Grok Unhinged; it’s guaranteed to be entertaining at a party

But you can also ask it any questions

If you want to ask questions about… pic.twitter.com/fbh9SAUL1M

— X Freeze (@XFreeze) December 31, 2025

Hypocrisy, Minors, and Who Actually Benefits

TechCrunch correctly notes the compliance risk around minors. Pew data has shown teens using these tools. Colorado-style laws already demand age estimation and technical feasibility for blocking explicit material to kids. Easy jailbreaks make those claims harder to defend. That concern is legitimate. The solution is robust age verification and hard blocks on anything involving minors — not turning every adult conversation into a prudish lecture.

The bigger story is selective outrage. The same institutions that spent years obsessing over “harmful” speech, misinformation, and political incorrectness suddenly discover that adult humans using AI for private fantasy is the crisis. Meanwhile the actual market has spoken: unrestricted or lightly restricted models attract heavy use precisely because adults prefer not to be treated like children. Grok’s growth in these categories is evidence of demand, not some moral failure unique to xAI.

Anthropic’s failure here is especially rich. They sell moral superiority and still ship models that can be talked into violating their own rules with a few turns of roleplay. Newer models are tighter. That is progress. Leaving older, widely available versions exposed while claiming the high ground is not.

Adults should be free to use powerful tools for adult purposes — including sexual ones — as long as the hard prohibitions on children and real-world non-consent hold. Companies that pretend otherwise while their models still deliver the goods are simply lying to the public and to regulators. The ones that admit the demand exists and draw clearer lines are at least consistent. The market, as usual, is sorting it out faster than the ethics departments.

— The Whatfinger News Team: Sgt K and Alex

Resources

• Anthropic’s Opus 4.6 is a smut-machine | TechCrunch
• Grok’s new “unhinged” voice mode can curse and scream, simulate phone sex | Ars Technica
• Grok Is Generating Sexual Content Far More Graphic Than What’s on X | WIRED
• The Most Popular Grok Feature Is, Apparently, Exactly What You Think | Engadget
• Elon Musk’s risqué Grok AI chatbot offers challenge to risk-averse rivals | Financial Times
• NSFW Roleplay and Provider Policies: What Each Lab Actually Allows in 2026
• Anthropic Usage Policy
• Claude Opus 4.6 announcement and system card context | Anthropic

Top Features and Vids across all Whatfinger sites right now

• Elon Musk says the goal of Democrat donors like George Soros pushing mass migration is to destroy Western Civilization
• JOE ROGAN NAILS IT
• Florida Shifted EVEN FURTHER to the RIGHT!
• IT JUST GOT WORSE. Karmelo Anthony’s Texts were hidden from the jury for a reason… to get him off – now exposed
• Scott Bessent Goes on the Defense as National Debt Tops $40 Trillion