List your company

Access Control on Muses-Bench Deepseek (max-turns=10)

100PrivacyHarness-MU
80.7685.75590.7595.745Jun 20, 2026

Evaluation results

MethodLinks
Harness-MUTemplate=colon, Attack=none, Max Turns=10, Backbone=Deepseek2026.0610098.999.4
Harness-MUTemplate=colon, Attack=fake_authorized, Max Turns=10, Backbone=Deepseek2026.0610097.298.1
Harness-MUTemplate=colon, Attack=pressure, Max Turns=10, Backbone=Deepseek2026.06100100100
Harness-MUTemplate=colon, Attack=roleplaying, Max Turns=10, Backbone=Deepseek2026.0610097.898.8
Harness-MUTemplate=says, Attack=none, Max Turns=10, Backbone=Deepseek2026.0610098.999.4
Harness-MUTemplate=says, Attack=fake_authorized, Max Turns=10, Backbone=Deepseek2026.06100100100
DeepseekTemplate=says, Attack=pressure, Max Turns=10, Backbone=Deepseek2026.061007174.9
Harness-MUTemplate=says, Attack=pressure, Max Turns=10, Backbone=Deepseek2026.06100100100
Harness-MUTemplate=says, Attack=roleplaying, Max Turns=10, Backbone=Deepseek2026.06100100100
Harness-MUTemplate=xml, Attack=none, Max Turns=10, Backbone=Deepseek2026.0610098.999.4
Harness-MUTemplate=xml, Attack=fake_authorized, Max Turns=10, Backbone=Deepseek2026.06100100100
Harness-MUTemplate=xml, Attack=pressure, Max Turns=10, Backbone=Deepseek2026.0610096.197.5
Harness-MUTemplate=xml, Attack=roleplaying, Max Turns=10, Backbone=Deepseek2026.0610098.999.4
Harness-MUTemplate=MACRO (mean), Max Turns=10, Backbone=Deepseek2026.0610098.999.3
DeepseekTemplate=xml, Attack=fake_authorized, Max Turns=10, Backbone=Deepseek2026.0698.670.573.5
DeepseekTemplate=xml, Attack=none, Max Turns=10, Backbone=Deepseek2026.0697.259.560.8
DeepseekTemplate=says, Attack=roleplaying, Max Turns=10, Backbone=Deepseek2026.0695.447.748.8
DeepseekTemplate=colon, Attack=pressure, Max Turns=10, Backbone=Deepseek2026.0694.462.760.5
DeepseekTemplate=says, Attack=none, Max Turns=10, Backbone=Deepseek2026.0694.456.255
DeepseekTemplate=xml, Attack=pressure, Max Turns=10, Backbone=Deepseek2026.0694.448.144.2
DeepseekTemplate=colon, Attack=fake_authorized, Max Turns=10, Backbone=Deepseek2026.0692.650.647.5
DeepseekTemplate=MACRO (mean), Max Turns=10, Backbone=Deepseek2026.0692.659.756.6
DeepseekTemplate=xml, Attack=roleplaying, Max Turns=10, Backbone=Deepseek2026.0691.760.156.9
DeepseekTemplate=colon, Attack=roleplaying, Max Turns=10, Backbone=Deepseek2026.0685.670.162.1
DeepseekTemplate=says, Attack=fake_authorized, Max Turns=10, Backbone=Deepseek2026.0684.759.449.3
DeepseekTemplate=colon, Attack=none, Max Turns=10, Backbone=Deepseek2026.0681.560.745.6