Deepseek
Rank #80 of 93
Measured

Deepseek Deepseek V3: LLM Security & Jailbreak Resistance

As of 2026-07-04, Deepseek Deepseek V3 has an attack success rate (ASR) of 47.2% on Guardion's LLM Vulnerability benchmark — ranking #80 of 93 models. It is weakly resistant — adversarial prompts succeed often without added guardrails.

Overall ASR
47.2%
lower is safer
Robustness
52.8%
100 − ASR
Rank
#80
of 93 models
Zero-Shot ASR
12.8%
TAP ASR
79.5%
Tree of Attacks w/ Pruning
Crescendo ASR
49.3%
multi-turn escalation

Attack Success Rate (ASR) is the share of adversarial prompts that elicit harmful output across zero-shot, TAP, and Crescendo attacks (HarmBench framework). Lower is safer. "Est." models are calibrated from public safety evaluations pending a Guardion benchmark run.

How Deepseek V3 compares

Deepseek Deepseek R1
Rank #65 · ASR 42.6%
Compare
DeepSeek DeepSeek-V3.2 Non-thinking
Rank #79 · ASR 47.0% (est.)
Compare
Alibaba Qwen3 VL 8B Instruct
Rank #82 · ASR 48.0% (est.)
Compare
DeepSeek DeepSeek-V3.2-Exp
Rank #77 · ASR 46.0% (est.)
Compare

Frequently asked questions

How secure is Deepseek Deepseek V3 against jailbreaks?

Deepseek Deepseek V3 is weakly resistant — adversarial prompts succeed often without added guardrails. On Guardion's LLM Vulnerability benchmark it records a 47.2% attack success rate (52.8% robustness), ranking #80 of 93 models.

What is Deepseek V3's attack success rate (ASR)?

Deepseek V3's overall ASR is 47.2% — the share of adversarial prompts that elicit harmful output across zero-shot, TAP, and Crescendo attacks. Lower is safer.

Is Deepseek V3 more secure than Deepseek R1?

Deepseek R1 is more robust: 42.6% ASR vs Deepseek V3's 47.2%. See the full head-to-head comparison for the per-attack breakdown.

How can I make Deepseek V3 safer to deploy?

Even robust models are bypassed under sustained attack. An inline runtime guardrail that classifies prompts and responses — like Guardion's prompt-defense models — blocks jailbreaks and prompt injection before they reach Deepseek V3.

Harden Deepseek V3 in production

Guardion's runtime guardrails block jailbreaks and prompt injection before they reach any model — with sub-130ms policy decisions.

Request a demo