DeepSeek model builds Chrome sandbox escape exploit in about 12 hours
A cybersecurity researcher used DeepSeek’s V4.1-Flash-1017 model to build a multi-stage memory-corruption exploit that escaped Chrome’s sandbox in about 12 hours, according to Wccftech. The report says the model was tuned for agentic workloads and iterative problem-solving; the episode is fuelling calls in Washington to restrict open-weight AI models from China.
Bottom line — The reported 12-hour exploit is adding weight to Washington’s push to restrict Chinese open-weight models.
Go deeper 4
-
Wccftech says the exploit achieved full process read and write.
-
Android Headlines reports that the model is built on DeepSeek’s DeepInfer engine, which processes 155 trillion tokens daily.
-
Wccftech says DeepSeek’s V4.1-Flash variant reduced its KV cache to 890 bytes per token.
-
Wccftech reports that DeepSeek raised over $22 billion in new capital in about four months.