Multi-turn attacks broke AI models 88% of the time — single-turn testing missed it, Cisco AI security lead warns at VB Transform 2026
When Cisco ran 6,986 multi-turn attacks against 15 flagship models , attackers who adapted across the conversation broke through as often as 88.3% of the time. Amy Chang, Cisco's head of AI threat int
When Cisco ran 6,986 multi-turn attacks against 15 flagship models , attackers who adapted across the conversation broke through as often as 88.3% of
Read Full Story at VentureBeat →Why This Matters
The high success rate of multi-turn attacks against AI models highlights a critical vulnerability in current AI security frameworks. As conversational AI systems become increasingly integrated into various sectors, the ability for attackers to exploit these weaknesses poses significant risks to data integrity and user trust.
Background Context
AI models have been rapidly adopted across industries, often without comprehensive security measures tailored to their unique functionalities. Historically, cybersecurity efforts have primarily focused on single-turn interactions, leading to a lack of preparedness against more sophisticated, multi-turn conversational strategies employed by adversaries.
What Happens Next
Organizations will need to reevaluate their AI security protocols and invest in more robust testing methodologies that account for the complexities of multi-turn interactions. This may lead to the development of new frameworks and tools specifically designed to anticipate and mitigate these advanced attack strategies.
Bigger Picture
The findings reflect a broader trend in cybersecurity, where the evolution of attack methods is outpacing defensive strategies. As AI technologies continue to advance, the challenge will be to create adaptive security solutions that evolve in tandem with the threats they face, ensuring a proactive rather than reactive approach to AI safety.
