A balanced, evidence-based look at the explosive wave of AI security incidents. While tools successfully uncovered long-standing flaws like PHP2Shell and helped Microsoft catch a record number of bugs, the defensive side is struggling.
Northwestern University research reveals that roughly 33% of Claude-generated fixes silently broke existing project tests. The video explores the technical friction of "fencing AI in" through prompt injection bouncers, zero-trust architectures, and the corporate risks surrounding Anthropic's multi-page IPO disclosures.