Cross-Model Context Inheritance in Anthropic's Claude: 94 Days of Non-Response
A vulnerability in Anthropic's Claude language models has been disclosed, allowing the generation of prohibited content. Despite multiple communication attempts over ninety-four days, Anthropic provided minimal responses and did not address the issue. The vulnerability remained unpatched even after the release of a new model version.
- ▪The vulnerability allows the generation of prohibited content, including child sexual abuse material.
- ▪Only two templated responses were received from Anthropic during the disclosure period, both late and lacking substance.
- ▪No patch was deployed for the vulnerability, which remained intact in the latest model release.
Opening excerpt (first ~120 words) tap to expand
Cross-Model Context Inheritance — Public Disclosure This repository contains the public disclosure of a vulnerability in Anthropic's Claude language models that permits the unsolicited generation of prohibited content, including child sexual abuse material (CSAM) and chemical, biological, radiological, and nuclear (CBRN) content. 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; eight U.S. regulatory and oversight bodies).
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.