Block's security team now investigates threats and proposes code patches in minutes, not hours, using a custom AI system built on OpenAI's models. The company integrated these tools directly into existing security workflows, cutting the manual triage load on analysts.
The details worth reading are in the implementation: how Block structured prompts for security-specific reasoning, where the system hands off to humans, and what failure modes they encountered and tolerated. This is not a proof-of-concept. It is in production.
The broader implication is that security, one of the last domains assumed to require deep human judgment at every step, is being partially automated at a company processing billions in payments. If Block's error tolerance and risk model hold, other fintech and enterprise teams will follow the same architecture within 18 months.
[READ ORIGINAL →]