June 17, 2026
AI Security Goes Beyond Prompt Injection
Audio version
AI systems create attack surfaces that traditional application security does not fully cover.
In Episode 14 of System Prompt, Val and Peter examine the security risks surrounding large language models, enterprise AI applications, tools, agents, and external integrations.
The conversation covers prompt injection, jailbreaks, system prompt extraction, context poisoning, supply-chain attacks, sensitive information disclosure, and poisoning through tools and Model Context Protocol integrations.
The central point is that AI security is not solved by filtering a few bad prompts.
Reliable protection requires layered controls, observability, instruction hierarchy, testing, limited permissions, trusted data sources, and clear boundaries around what models and tools are allowed to do.
WHAT WE DISCUSS
• Why AI expands the application attack surface
• Direct and in