Imposter AI procedures produce a 37.50% weighted single-turn ASR, Soft Paraphrase produces 29.21%, and System Prompts produce 27.69%.
Imposter AI procedures produce a 37.50% weighted single-turn ASR, Soft Paraphrase produces 29.21%, and System Prompts produce 27.69%. — This cybersecurity statistic was published by Cisco in May 2026. It covers topics including Adversarial Attacks, Attack Techniques, Model Safety. The original data appears in Proprietary Problems: How Frontier Closed Models Collapse Under Iterative Pressure. For the full methodology and detailed findings, refer to the original report.
Share or Copy this stat
Frequently Asked Questions
What does this statistic say?
Imposter AI procedures produce a 37.50% weighted single-turn ASR, Soft Paraphrase produces 29.21%, and System Prompts produce 27.69%. This data was published by Cisco and covers Adversarial Attacks, Attack Techniques, Model Safety.
Where does this data come from?
This statistic comes from Proprietary Problems: How Frontier Closed Models Collapse Under Iterative Pressure, published by Cisco on May 31, 2026. You can view the original report at https://www.cisco.com/content/dam/cisco-cdc/site/en_us/products/security/proprietary_problems.pdf.
What cybersecurity topics does this cover?
This statistic relates to Adversarial Attacks, Attack Techniques, Model Safety. Browse more statistics on Adversarial Attacks or from Cisco.