Beyond the Hype The Real Role of Red Teams in an AI World

The session opens with adversary simulation, the core work of red teams. Because the threat landscape is too large to cover, teams must pick which threat actors to emulate, and the speakers describe a trade-off: higher fidelity to a known actor costs more time and effort, while emulating several actors at once sacrifices accuracy. They suggest grounding these choices in threat intelligence and treating costly capabilities, such as developing zero-day exploits, as an ongoing research effort rather than a single operation.

Next, they argue that red teams drive change by telling a clear story tied to business impact and by acting as a sparring partner for defenders. Sharing the full activity log, including failed attempts, helps blue teams see where to add or fix detection signals. Engagements can also surface known unknowns, risks people suspect but cannot act on, including the watermelon problem where indicators look green but the system is red.

On AI, the speakers say attackers gain scale, speed and tireless patience, so defenders need faster benchmarking, tests of existing controls against AI-driven intrusion, and new controls such as honey tokens. Mapping exercises into attack graphs can reveal choke points worth investing in, and AI agents themselves introduce prompt-injection risk. Their takeaway is that AI changes attackers, defenders and red teams alike, and red teams should use it to emulate and study threats.