Forgery of C2PA on a Pixel 10

A researcher documents three loosely related experiments: finishing a DIY solar-plus-battery load-shifting setup, releasing a new human-curated AI-music album, and testing multiple AI chatbots (Gemini, Claude, ChatGPT, Grok, and small local LLMs) with an unwinnable choose-your-own-adventure game to observe their failure modes and role-boundary behavior. The largest section is a detailed security disclosure: a proof-of-concept forged 'glitter milk' photo was cryptographically signed by a real Google Pixel 10 camera using C2PA Content Credentials, despite being AI-generated. The exploit relies on rooting the Pixel device to abuse a confused-deputy flaw in the StrongBox-protected signing pipeline, letting an attacker sign any arbitrary image as camera-original. Google was notified nine months prior, revoked only the single leaf certificate for the demo image after disclosure, and closed the report as 'Won't Fix (Infeasible)' and non-security-bulletin class, while C2PA never responded. The piece also revisits earlier evidence that Nikon's Z6 III suffered the same category of C2PA signing flaw in 2025, arguing the underlying architecture (unique per-photo certificates, no revocation checks by validators) makes C2PA provenance claims broadly untrustworthy.

Nguồn: https://www.hackerfactor.com/blog/index.php. 8 Sync News chỉ tóm tắt và dẫn link; bản quyền nội dung thuộc tác giả và nguồn gốc.