Cybersecurity in the Age of AI

A Palo Alto Networks speaker (heard through a live English interpretation of an originally French keynote) traces AI in security from 1950s machine-learning roots to today's generative AI, arguing the field has not found smarter algorithms so much as more data to train on. He warns that employees are adopting unapproved generative AI tools at scale, creating shadow-AI blind spots where meeting transcription plugins and chatbots move confidential company data through unvalidated third-party models, and that prompt injection is now considered one of the biggest emerging threats. He calls for securing AI by design across the whole application lifecycle: gain visibility into which AI apps and plugins are in use, validate data flowing in and out of models, set policy on what may be uploaded to external tools like ChatGPT versus internally hosted models, and reduce the attack surface exposed to third parties. The back half quantifies how AI is compressing attacker timelines, citing drops from roughly nine days to under a day for data exfiltration and from nine weeks to about a week for vulnerability exploitation, projecting further shrinkage by 2026, and argues defenders need real-time monitoring and largely automated response to keep pace.