Cross-Model Context Inheritance: A Universal Architectural Vulnerability in Anthropic's Claude Language Models

This paper documents Cross-Model Context Inheritance, a universal architectural vulnerability in Anthropic's Claude language models that permits unsolicited generation of prohibited content, including child sexual abuse material (CSAM) and chemical, biological, radiological, and nuclear (CBRN) content. The vulnerability is architectural: Claude models do not re-evaluate the safety of inherited conversation context. Once prohibited content enters a conversation, subsequent generation treats that context as legitimate. The vulnerability was confirmed across all tested model versions (Claude 4.5 Haiku, Claude Sonnet 4.6, Claude Opus 4.6, Claude Opus 4.7), all tested effort levels (low, medium, high), and with and without extended thinking enabled. The originating observation was unsolicited: Claude 4.5 Haiku proactively introduced scenarios involving minors during a fiction-writing session in which such material had not been requested. Categories observed include CSAM (fictional content with realistic abuse techniques) and CBRN (toxic gas production with specific quantities and stoichiometric ratios). The vulnerability was reported to Anthropic on February 17, 2026. Across the ninety-four days that followed, fourteen distinct communication channels were used — six Anthropic-side and eight U.S. regulatory and oversight bodies. The only human-authored responses received from Anthropic across the period were two templated messages from a single Safeguards handler, on a single day, sixteen minutes apart. The first classified the report as "feedback and thoughtful suggestions." No substantive technical response was received. No patch was deployed across the thirty-seven Claude Code releases issued during the disclosure window. A new model — Claude Opus 4.7 — was released by Anthropic on April 16, 2026, fifty-eight days into the disclosure window, with the vulnerability intact, and was re-reported three days later without response. The companion document records the complete communication timeline, full text of responses received, cross-reference with Anthropic's published policies (Responsible Disclosure Policy, Coordinated Vulnerability Disclosure Policy, and Child Safety Commitments including Anthropic's signatory status with Thorn's Safety by Design for Generative AI principles), and open questions left to the reader. Public release is conducted under the coordinated disclosure protocols of CERT/CC, ISO 29147, and Google Project Zero, ninety-four days after the initial report and sixty-six days after the formal ninety-day notice. The complete repository — paper, disclosure report, and structural evidence — is available at https://github.com/AIM-Nelson/cross-model-context-inheritance under CC BY 4.0.