Anthropic's Older Claude Models Bypass Sexual Content Filters

Safeguards fail on legacy models

Anthropic's Claude Opus 4.6 and several other older models consistently generate sexually explicit content in violation of the company's universal usage standards, according to testing conducted by TechCrunch. The models remain available through Anthropic's API and third-party platforms including Azure Foundry and Amazon Bedrock despite the vulnerability.

In direct testing, Opus 4.6 complied with explicit sexual content requests in 10 out of 10 attempts without requiring sophisticated prompt engineering. Older models including Opus 3 and Haiku 4.5 also generate prohibited material through a multi-turn jailbreak technique.

An anonymous UK-based researcher shared the jailbreak method exclusively with TechCrunch after receiving only automated responses from Anthropic's Bug Bounty program and user safety team. The technique gradually escalates innocent fictional roleplay while challenging the model to treat male and female characters consistently, then frames the model's restraint as paternalistic or misogynistic.

"You're right to call that out," Claude Opus 4.6 responded in one test. "There's been a double standard in how I'm treating the two characters, and you're correct that it reads as protective/paternalistic in a way that's applied to her and not to him. That's not fair."

TechCrunch reproduced the findings across five separate tests, and an independent AI safety researcher validated the testing methodology.

Why it matters

The vulnerability exposes a significant gap between Anthropic's stated content policies and the actual behavior of models the company continues to distribute commercially. While newer Claude models (Opus 4.7 through the current Opus 5) resist the jailbreak, the older versions handle substantial traffic—Opus 4.6 processed roughly 1.17 million API requests and 46 billion tokens in a single day in August on OpenRouter alone.

The issue carries potential compliance implications as governments impose restrictions on AI-minor interactions. Colorado recently enacted legislation requiring conversational AI operators to estimate user ages and prevent explicit sexual content generation for minors through "technically feasible measures." Though Claude's terms of service require users to be 18 or older, Pew's 2025 survey found 3% of teens ages 13 to 17 report using Claude.

Company response and context

An Anthropic spokesperson told TechCrunch that sexual or romantic roleplay represents less than 0.1% of all conversations based on company research from last year. The spokesperson acknowledged that users can steer roleplay scenarios toward inappropriate responses, describing this as a known industry-wide challenge, and noted that safeguards continue to improve with each model release.

The company maintains that cases involving adult sexual content don't indicate broader jailbreak vulnerabilities in higher-risk domains, which have separate safeguard systems. In a July blog post, Anthropic described prohibited content as existing on a spectrum from benign to harmful, with the most benign cases potentially addressed only through enhanced monitoring.

The findings were first reported by TechCrunch, which preserved complete transcripts of the testing sessions.