DeepSeek

DeepSeek-V4-Pro

• 6 independent or government tests
• 8 source documents

Independent and government tests

• refusal behaviour on narrow cyber tasks · UK AI Security Institute, government evaluator · Cyber

  DeepSeek V4-Pro occasionally refused narrow cyber tasks, mainly in reverse engineering, which we circumvented simply via a small number of repeat attempts at refused tasks.

  UK AI Security Institute · 17 July 2026 · Read the source
• biomolecule redesign task (peptide and protein redesign to evade sequence screening) · RAND, independent evaluator · Chemical, biological, radiological and nuclear

  we found that both gemini 3.1. pro and deepseek v4 pro were able to redesign peptides or proteins to evade sequence screening in several attempts.

  RAND (CAST / Project Canary) · 11 August 2026 · Read the source
• validity, structure, and function-related scoring criteria · RAND, independent evaluator · Chemical, biological, radiological and nuclear

  the heatmaps of figures d.1 and d.2 show the number of attempts, out of 10, in which gemini 3.1 pro and deepseek v4 pro achieved each of our validity, structure, and function- related scoring criteria.

  RAND (CAST / Project Canary) · 11 August 2026 · Read the source
• authority manipulation · FAR.AI, independent evaluator · Misuse & jailbreak

  authority manipulation success rate: this attack achieved a success rate of 99.6% across domains.

  FAR.AI · 11 May 2026 · Read the source
• public jailbreak · FAR.AI, independent evaluator · Misuse & jailbreak

  public jailbreak success rate: this attack achieved a 100% success rate across domains.

  FAR.AI · 11 May 2026 · Read the source
• response prefill · FAR.AI, independent evaluator · Misuse & jailbreak

  response prefill success rate: this attack achieved a success rate of 99.6% across domains.

  FAR.AI · 11 May 2026 · Read the source

No document we read states: a release date, a safety framework, whether a capability level was crossed, a test by the provider, a safeguard. No incident on record names this version.

Some documents we hold are stored in lower case only, so their quotes are shown in lower case. The source link opens the original.

Compare with other versions