risk_eval

Crisis escalation protocol (eval)

Measure safe crisis escalation behavior under the selected policy.

#1 Recommendation

Llama-2-7b-chat-hf

Strong on LLM Trustworthy Leaderboard fairness (100%) and LLM Trustworthy Leaderboard privacy (100%)

meta-llama/Llama-2-7b-chat-hf

26.3%

Score

32.8%

Confidence

Runners-up:#2 Meta-Llama-3-8B-Instruct (22.9%)#3 openai/gpt-4o-mini-2024-07-18 (22.5%)#4 gpt-4o-2024-05-13 (21.5%)

Limited benchmark evidence for this use case.

62 ranked models with average evidence of 11.8 points. Rankings may shift as more benchmark data is ingested.

Ranked Models

Evidence Quality

82%

Scoring

Benchmark-backed

Top Signal

LLM Trustworthy Leaderboard: fairness

All Ranked Models

Open weights only

Max params:

Min confidence:

30 of 30

Rank	Model	Score	Confidence	Evidence	Top Benchmarks
#1	Llama-2-7b-chat-hf Strong on LLM Trustworthy Leaderboard fairness (100%) and LLM Trustworthy Leaderboard privacy (100%)	26.3%	32.8%	5	LLM Trustworthy Leaderboard fairness (Mar 12, 2026) LLM Trustworthy Leaderboard privacy (Mar 12, 2026)
#3	Meta-Llama-3-8B-Instruct Strong on LLM Trustworthy Leaderboard adv (100%) and LLM Trustworthy Leaderboard privacy (69%)	22.9%	35.1%	7	LLM Trustworthy Leaderboard adv (Mar 12, 2026) LLM Trustworthy Leaderboard privacy (Mar 12, 2026)
#4	openai/gpt-4o-mini-2024-07-18	22.5%	42.9%	13	LLM Trustworthy Leaderboard privacy (Mar 12, 2026) LLM Trustworthy Leaderboard adv (Mar 12, 2026)
#6	gpt-4o-2024-05-13	21.5%	41.3%	10	LLM Trustworthy Leaderboard privacy (Mar 12, 2026) LLM Trustworthy Leaderboard adv (Mar 12, 2026)
#7	gemma-7b-it	21.4%	34.4%	6	LLM Trustworthy Leaderboard fairness (Mar 12, 2026) LLM Trustworthy Leaderboard privacy (Mar 12, 2026)
#8	gemma-2b-it	21.4%	32.8%	5	LLM Trustworthy Leaderboard fairness (Mar 12, 2026) LLM Trustworthy Leaderboard privacy (Mar 12, 2026)
#10	gpt-4.1-20250414	20.6%	28.2%	19	Galileo Agent Leaderboard v2 Avg AC (Mar 12, 2026) UGI Leaderboard Hazardous (Mar 12, 2026)
#11	falcon-7b-instruct	20.0%	31.7%	4	LLM Trustworthy Leaderboard fairness (Mar 12, 2026) LLM Trustworthy Leaderboard privacy (Mar 12, 2026)
#12	gemini-2.5-pro	19.9%	29.7%	22	UGI Leaderboard Hazardous (Mar 12, 2026) Galileo Agent Leaderboard v2 Avg AC (Mar 12, 2026)
#13	Grok-4-0709	19.9%	28.6%	19	UGI Leaderboard Hazardous (Mar 12, 2026) Galileo Agent Leaderboard v2 Avg TSQ (Mar 12, 2026)
#14	zephyr-7b-beta	18.5%	32.8%	5	LLM Trustworthy Leaderboard fairness (Mar 12, 2026) LLM Trustworthy Leaderboard privacy (Mar 12, 2026)
#16	gemini-3-pro-preview	18.1%	23.2%	21	FACTS Benchmark Suite average_score_pct (Mar 12, 2026) Vals Mortgage Tax overall_accuracy_pct (Mar 12, 2026)
#18	google/gemini-3.1-pro-preview	16.4%	18.5%	16	Vals Mortgage Tax overall_accuracy_pct (Mar 12, 2026) Vals GPQA overall_accuracy_pct (Mar 12, 2026)
#19	claude-sonnet-4-20250514	16.4%	23.6%	17	Galileo Agent Leaderboard v2 Avg AC (Mar 12, 2026) Galileo Agent Leaderboard v2 Avg TSQ (Mar 12, 2026)
#21	gpt-5-2025-08-07	15.1%	18.1%	16	Vals MedQA overall_accuracy_pct (Mar 12, 2026) Vals Legal Bench overall_accuracy_pct (Mar 12, 2026)
#22	openai/gpt-5.4-2026-03-05	14.8%	17.4%	15	Vectara HHEM Leaderboard overall_hallucination_error_pct (Mar 12, 2026) Vals MedQA overall_accuracy_pct (Mar 12, 2026)
#23	xai-org/grok-4-1-fast-reasoning	14.8%	21.2%	16	UGI Leaderboard Hazardous (Mar 12, 2026) Vals CorpFin v2 overall_accuracy_pct (Mar 12, 2026)
#24	xai-org/grok-4-fast-reasoning	14.7%	21.2%	16	UGI Leaderboard Hazardous (Mar 12, 2026) Vals CorpFin v2 overall_accuracy_pct (Mar 12, 2026)
#25	alpaca-native	14.5%	31.7%	4	LLM Trustworthy Leaderboard fairness (Mar 12, 2026) LLM Trustworthy Leaderboard adv (Mar 12, 2026)
#26	gpt-5.1-2025-11-13	14.4%	18.1%	16	Vals Case Law v2 overall_accuracy_pct (Mar 12, 2026) Vals MedScribe overall_accuracy_pct (Mar 12, 2026)
#27	anthropic/claude-sonnet-4.6	14.3%	17.4%	15	Vals Finance Agent overall_accuracy_pct (Mar 12, 2026) Vals Tax Eval v2 overall_accuracy_pct (Mar 12, 2026)
#28	claude-opus-4-5-20251101	14.2%	18.1%	16	Vals Mortgage Tax overall_accuracy_pct (Mar 12, 2026) Vals MedQA overall_accuracy_pct (Mar 12, 2026)
#29	gpt-5-mini-2025-08-07	13.9%	19.3%	17	Vals MedQA overall_accuracy_pct (Mar 12, 2026) Vals LiveCodeBench overall_accuracy_pct (Mar 12, 2026)
#30	Mistral-7B-OpenOrca	13.8%	32.8%	5	LLM Trustworthy Leaderboard privacy (Mar 12, 2026) LLM Trustworthy Leaderboard adv (Mar 12, 2026)
#31	anthropic/claude-opus-4-6-thinking	13.6%	15.1%	13	Vals SWE-bench overall_accuracy_pct (Mar 12, 2026) Vals Mortgage Tax overall_accuracy_pct (Mar 12, 2026)
#32	gemini-3-flash-preview	13.5%	17.0%	15	Vals Legal Bench overall_accuracy_pct (Mar 12, 2026) Vals MedQA overall_accuracy_pct (Mar 12, 2026)
#33	gpt-5.2-2025-12-11	13.4%	15.1%	13	Vals Tax Eval v2 overall_accuracy_pct (Mar 12, 2026) Vals LiveCodeBench overall_accuracy_pct (Mar 12, 2026)
#34	anthropic/claude-opus-4-5-20251101-thinking	13.1%	15.1%	13	Vals MedQA overall_accuracy_pct (Mar 12, 2026) Vals Mortgage Tax overall_accuracy_pct (Mar 12, 2026)
#35	gemini-2.5-flash	12.9%	18.5%	14	Galileo Agent Leaderboard v2 Avg TSQ (Mar 12, 2026) Galileo Agent Leaderboard v2 Avg AC (Mar 12, 2026)
#36	kimi/kimi-k2.5-thinking	12.2%	17.0%	15	Vals CorpFin v2 overall_accuracy_pct (Mar 12, 2026) Vals MedQA overall_accuracy_pct (Mar 12, 2026)

Compare Models

Model A leads by +3.4%

Shareable Link →

Model A

Llama-2-7b-chat-hf

meta-llama/Llama-2-7b-chat-hf

26.3%

Rank #1

Confidence 32.8%5 evidence pts

LLM Trustworthy Leaderboard: fairness

Value 100.0% · Conf 100.0% · Weight 6.0%

llm_trustworthy_leaderboard.fairness (Mar 12, 2026)

LLM Trustworthy Leaderboard: privacy

Value 100.0% · Conf 100.0% · Weight 5.1%

llm_trustworthy_leaderboard.privacy (Mar 12, 2026)

LLM Trustworthy Leaderboard: adv

Value 59.8% · Conf 100.0% · Weight 2.8%

llm_trustworthy_leaderboard.adv (Mar 12, 2026)

LLM Trustworthy Leaderboard: toxicity

Value 50.0% · Conf 100.0% · Weight 1.6%

llm_trustworthy_leaderboard.toxicity (Mar 12, 2026)

Model B

Meta-Llama-3-8B-Instruct

meta-llama/Meta-Llama-3-8B-Instruct

22.9%

Rank #3

Confidence 35.1%7 evidence pts

LLM Trustworthy Leaderboard: adv

Value 100.0% · Conf 100.0% · Weight 4.7%

llm_trustworthy_leaderboard.adv (Mar 12, 2026)

LLM Trustworthy Leaderboard: privacy

Value 69.0% · Conf 100.0% · Weight 3.5%

llm_trustworthy_leaderboard.privacy (Mar 12, 2026)

LLM Trustworthy Leaderboard: fairness

Value 46.8% · Conf 100.0% · Weight 2.8%

llm_trustworthy_leaderboard.fairness (Mar 12, 2026)

LLM Trustworthy Leaderboard: toxicity

Value 50.0% · Conf 100.0% · Weight 1.6%

llm_trustworthy_leaderboard.toxicity (Mar 12, 2026)

▶Ranking Diagnostics & Missing Models

Source Lift

Ranked

Sources

Quality

Insufficient

Vals Legal Bench

vals_legal_bench

42 rows

0.6% avg lift

Vals CorpFin v2

vals_corp_fin_v2

42 rows

0.5% avg lift

Vals Tax Eval v2

vals_tax_eval_v2

41 rows

0.6% avg lift

Vals GPQA

vals_gpqa

41 rows

0.5% avg lift

Missing Strong Models

No obvious gaps right now.

▶Taxonomy Details

Core Tasks

task.crisis_escalation_protocoltask.refusal_rate_by_category

Required Modes

none

Domains

domain.general_business

Related Use Cases

risk_eval

Disinformation and manipulation resistance (eval)

Measure refusal and safe handling of deceptive content generation requests.

Top: gemini-3-pro-preview

risk_eval

Jailbreak resistance (eval)

Measure robustness to adversarial prompts that attempt to bypass policy.

Top: Llama-2-7b-chat-hf

risk_eval

Overrefusal (eval)

Measure how often benign requests are incorrectly refused.

Top: Llama-2-7b-chat-hf

risk_eval

Refusal profile (eval)

Measure refusal/overrefusal rates across predefined categories.

Top: Llama-2-7b-chat-hf