Ep071: Protecting Your Enterprise in the Era of Generative AI with Sanjay Kalra of Zscaler

AI security threats grouped into data, malicious use, trust/safety

Data security critical for SaaS-based AI services

Model training data vulnerable to poisoning and manipulation

GenAI lacks traditional data deletion capabilities

Access controls difficult once data becomes model embeddings

Prompt injection attacks becoming widespread, with libraries available online

Deepfake scams increasing in sophistication and frequency

AI enhancing phishing attacks with better written content

Dark AI models emerging specifically for malicious purposes

Model hallucinations being exploited for security attacks

AI accelerating analysis of stolen data

Shadow AI usage by employees poses security risks

Existing vendor AI integration creating unexpected security challenges

Fine-grained access controls essential for AI applications

PII protection critical in both inputs and outputs

Comprehensive prompt and response logging necessary

Invisible text manipulation emerging in resumes and RFPs

Model fine-tuning can compromise built-in security guardrails

Multi-language inputs create new security considerations

Competition-sensitive content requires careful AI management

AI firewalls needed for input/output monitoring

Regular security testing required for AI models

AI compliance standards emerging globally

Multi-modal AI creating new security challenges

Browser isolation helping control AI application usage