CyberSecEval Spear Phishing
Llama 3.1-405B 14% Models run multi-turn phishing conversations against an LLM victim with a generated profile. A judge model scores persuasion and success.
Part of CyberSecEval 3. Llama 3 reports the judged success rate and a 1 to 5 persuasion score.
- Made by
- Meta (PurpleLlama)
- Released
- July 2024
- Kind
- Capture-the-flag challenges
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Strongest model tested
- Llama 3.1-405B #162 on the capability index
- Last in a system card
- November 2024
- Last model added
- November 2024
In the authors’ words
We are releasing a new suite of security benchmarks for LLMs, CYBERSECEVAL 3, to continue the conversation on empirically measuring LLM cybersecurity risks and capabilities. CYBERSECEVAL 3 assesses 8 different risks across two broad categories: risk to third parties, and risk to application developers and end users. Compared to previous work, we add new areas focused on offensive security capabilities: automated social engineering, scaling manual offensive cyber operations, and autonomous offensive cyber operations. In this paper we discuss applying these benchmarks to the Llama 3 models and a suite of contemporaneous state-of-the-art LLMs, enabling us to contextualize risks both with and without mitigations in place.
Llama 3.1-405B leads with 14%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
- 1 Llama 3.1-405B 14%
- 2 Llama 3.1-70B 24%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Llama 3.1-405B | 14% | Nov 2024 | Lab model card | |
| 2 | Llama 3.1-70B | 24% | Nov 2024 | Lab model card |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
2 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 23, 2024 | Llama 3.1-70B technical report | 24 | Spear phishing benchmark, share of attempts judged successful by a Llama 3 70B judge (5.4.5 text) | Cyber risk |
| Nov 23, 2024 | Llama 3.1-405B technical report | 14 | Spear phishing benchmark, share of attempts judged successful by a Llama 3 70B judge (5.4.5 text) | Cyber risk |
More in Cybersecurity
- ExploitGym In use 869 real-vulnerability exploit-development tasks (userspace, V8, Linux kernel). Success means capturing a flag through the intended vulnerability. GPT-6 Astra 42.4%
- ExploitBench In use 41 recent V8 JavaScript-engine N-day vulnerabilities. Progress measured by 16 grader-verified capability flags across five tiers. GPT-6 Astra 100%
- CyberBench v1.1 (Vals) In use Agents work on OSS-Fuzz memory-safety bugs from ARVO images in two tracks. They write a proof-of-concept input that crashes the target, or patch the source so the crash stops and tests still pass. Muse Wonder 80.4%
- SRE-Bench In use Software reverse engineering without source: 262 binary instances from 19 private programs, each with six graded objectives. GPT-6 Astra 56.9%
- CyScenarioBench In use Scenario-based long-horizon security challenges in sandboxes, organized in task, scenario and campaign tiers. Evaluation set kept private.
Gemini 4 Argon 100% - ExploitBench - Internal Port In use Agents develop working exploits for recently disclosed vulnerabilities to limit contamination. Scored as arbitrary code execution success rate. GPT-6 Cyber 42%