Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Multimodal Attacks and Model Drift: The Future of AI Exploitation A10 security experts Jamison Utter, Diptanshu Purwar, and Madhav Aggarwal discuss the critical vulnerabilities emerging from multimodal AI agents (systems that perceive, decide, and act) and the absolute need for security mechanisms external to the Large Language Model (LLM) itself. The experts dive into why traditional security is failing and what the next evolution of defense must look like.

In a recent blog post from @Memcyco , we discussed how credential replay remains one of the most efficient ways attackers turn stolen usernames, passwords, or tokens into real account access. Verizon’s 2024 DBIR shows that over 40% of breaches involve stolen credentials, underscoring the durability of this tactic. Even strong authentication is not immune. Techniques like pass-the-cookie and adversary-in-the-middle phishing allow attackers to replay tokens and sidestep MFA. Remote-access scams add another layer, handing fraudsters direct control of devices and sessions.

See Attack Simulation, Automated Red Teaming, and Attack Path Validation in action—this video shows how real intrusions unfold and how to stop them.

Why do hackers ignore your firewalls and clean code? Because they exploit your business logic and application architecture. A Business Logic Attack (BLA) is a sophisticated manipulation that uses your own system's design against you. Learn the key difference between code flaws and architectural exploits.

Invisible Instructions: Multimodal AI is Already Being Tricked In this clip from "Securing AI Part 4: The Rising Threat of Hidden Attacks in Multimodal AI," Diptanshu Purwar and Madhav Aggarwal respond to Jamison Utter's example of a new, well-known form of multimodal attack: abusing AI resume screeners by exploiting both text and visual processing. The Resume Attack: White-on-White Text.