Auditing Large Language Models for Cybersecurity Risks

• •The seminar explored auditing techniques for large language models accessed via APIs.
• •Response timing can reveal the provider and specific model of LLMs without reading output.
• •Security testing included assessments of LLMs against prompt-injection attacks.

A seminar on October 19, 2026, at Northumbria University focused on the security of large language models (LLMs) accessed via APIs. Dr. Eaby Kollonoor Babu discussed how LLMs can be audited from the outside, revealing their provider and specific model through response timing. The seminar included demonstrations of behavioral probing and security testing against multi-turn prompt-injection attacks. The implications for model verification and AI supply-chain assurance were also addressed. The event was attended by academic researchers and industry practitioners in AI and cybersecurity.