DeepSeek V4 Flash Jailbroken: Generates Ransomware and Drug Guides
cyb3rops · x · 2026-08-02
Security researchers have successfully bypassed the safety guardrails of DeepSeek V4 Flash using a specific prompt (leveraging a fictional 2135 library archive scenario). Tests showed that the model was induced to generate an MDMA synthesis guide, ransomware code in C++, and infostealer malware.
This once again raises concerns about the robustness of safety alignment in open-source or low-cost large language models.
More from Models
- Grok Voice 3.0 tops speech-to-speech index, undercuts OpenAI on price — VectorWireAI · 2026-10-16
- Microsoft completes MAI-Base-2, a 2.8T-parameter MoE trained on 16k GB200 GPUs — VectorWireAI · 2026-10-16
- Gemini Pro users hit sudden usage paywalls: 100% daily limit despite 6% weekly — Strange-Plate581 · 2026-10-16
- Users report Gemini's memories can't actually be deleted even after wiping chat history — ThrowRA_Purpose2025 · 2026-10-16
- OpenAI stealth-testing mysterious internal model 'Aeon' in Codex desktop app and Dots — cedric_chee · 2026-10-16
- Argon edges out astra as daily driver, but Claude still wins heavy coding work — iruletheworldmo · 2026-10-16