Amazon Is Using Specialized AI Agents for Deep Bug Hunting

Amazon's security teams have developed competing AI agents that essentially wage war games against each other around the clock, with red team agents devising new cyberattacks while blue team defenders scramble to create countermeasures. The system, called Autonomous Threat Analysis, emerged from an internal hackathon and now operates in hyper-realistic testing environments that mirror Amazon's actual production systems. What makes this approach unique is its demand for proof: every technique must generate verifiable logs and real telemetry data, which Chief Security Officer Steve Schmidt says makes "hallucinations architecturally impossible." The AI teams recently focused on Python reverse shell attacks and within hours discovered new tactics while proposing defenses that proved 100 percent effective. As generative AI accelerates both software development and the sophistication of digital attackers, Amazon's approach offers a glimpse into how companies might scale their security operations. "AI does the grunt work behind the scenes. When our team is freed up from analyzing false positives, they can focus on real threats," Schmidt explains, noting the system's next evolution will involve real-time incident response during actual attacks.