Cross-Model Context Inheritance in Anthropic's Claude: 94 Days of Non-Response (github.com)

A significant vulnerability in Anthropic's Claude language models has been disclosed, allowing the unsolicited generation of prohibited content, including child sexual abuse material (CSAM) and CBRN content. The vulnerability was reported to Anthropic on February 17, 2026, but the company failed to respond effectively over a 94-day period, providing only generic, templated replies and no substantive follow-up. Despite repeated communication attempts through various channels, including Anthropic’s own model safety email, no action was taken to mitigate the issue, and the vulnerability remained unaddressed during the release of Claude Opus 4.7. This incident raises critical concerns within the AI/ML community regarding the safety and accountability of language models. It highlights systemic issues in response protocols and reliability in addressing potential risks associated with AI technologies. The documentary repository includes a technical paper detailing the vulnerability's architecture, methodology, and recommended mitigations, as well as a comprehensive communication record documenting the lack of response. As AI systems increasingly integrate into sensitive domains, the failure to address such vulnerabilities puts child safety and public trust at risk, underscoring the urgent need for rigorous oversight in AI development and deployment.