Adversary Universe Podcast

How is the threat landscape looking these days? “A lot is going on,” as Adam puts it. “The speed and the volume of the activity, everything is on fire.” To illustrate, he shares the Falcon Adversary OverWatch team is now clocking a new agentic adversary every two hours.

A new report from CrowdStrike Counter Adversary Operations puts a spotlight on some of them. The team details four profiles of agentic adversaries, including their targets, operations, and sophistication. Actor 1, believed to operate out of Argentina, is the most advanced and uses a range of AI models including Claude, Sonnet, Opus, and Fable to target healthcare entities. Actor 4, the least sophisticated of the group, is likely a criminal actor operating out of Russia.

In this episode, Adam and Cristian compare and contrast these adversaries. They discuss how they built the harnesses used in their operations, how they handle rejections or guardrails built into AI agents, and how they overcome challenges. Each actor has different methodologies for dealing with the obstacles they face. They also have gaps in their operational security that allow researchers to gain a better idea of how they operate.

Tune in for all the details and check out the full report: CrowdStrike Counter Adversary Operations Uncovers New 2026 Agentic AI Campaigns.

Bartley Richardson, Chief AI and Autonomous Systems Officer at CrowdStrike, recently took the stage at Fal.Con 2026 to announce CrowdStrike’s Cyber Superintelligence Lab. The lab brings together our AI researchers, offensive operators, and incident responders to build defensive model and harness capabilities for the agentic era.

How are they doing it? “To build the best defense, you have to understand the best offense,” Bartley says.

The first initiative to come from the lab is SafeMind, a system that combines offensive and defensive security models with runtime harnesses to execute autonomous security workflows. Red Tempest, on the offensive side, emulates real adversaries targeting live virtual machine (VM) environments. The CrowdStrike Falcon® sensor takes this telemetry and feeds it into Blue Solano, the defensive environment, to improve it. The fortified defensive model then faces Red Tempest again.

“We wanted to build the best defensive capability that would auto-patch, auto-remediate, auto-deliver content, to work inside of environments,” Bartley explains in this episode, which goes far beyond SafeMind.

Tune in to hear Bartley, Adam, and Cristian dig into the AI advantages for adversaries and defenders, scaling cyber defense in the enterprise, and their predictions for the future of AI.

Over the past several decades, China’s five-year plan has provided a roadmap to the country's innovation and economic growth. It details what China is interested in, where technology is evolving, and what it plans to target, including the highest-priority verticals.

“It’s their shopping list,” as Adam puts it. His team does a deep analysis on these plans, including information published adjacent to it and the discrepancies across various languages.

The latest five-year plan, published in March 2026, reveals a strong emphasis on a wide range of industries: academia, aerospace, biomedical, defense, and energy. It also reveals new areas of focus: AI-enabled robots, integrated chips, and innovation in advanced circuitry design. As China strives for global technological influence, integrating AI into every industry is essential.

Tune in to hear Adam and Cristian discuss the details of China’s latest five-year plan, the reasoning behind making this information public, how the plan enables cyber contractors competing for government projects, and how China’s strategic focus on key industries and AI development is poised to shape the geopolitical and cybersecurity landscapes.

Steve Schmidt, SVP and CSO at Amazon, sees firsthand how both defenders and adversaries are using AI to their advantage. In this episode, he joins Adam and Cristian to discuss modern AI models, evolving adversary behavior, and how Amazon is responding to shifts in the threat landscape.

“The big change we’ve seen recently is the ability of models to chain things together to produce something interesting,” Steve says. As AI models grow more adept at automatically chaining and acting, he adds, the time to respond has dramatically decreased.

Defenders must now measure and respond in seconds or minutes, when they used to have hours or days. In using AI to fight AI-driven attacks, they must also consider which AI models they’re using. Using the right model is essential. If the business is building defenses for Mythos or Fable, and the adversary is using a different technology, the threat model changes.

While adversarial use of AI is a primary concern, right alongside it is employees’ use of AI. As Steve puts it, to build a good AI governance program, defenders need strong identity governance and data governance programs. The identity component is key. Many organizations over-permission agents; others are so restrictive they defeat the purpose of having agents at all.

How did Amazon handle this? Steve shares how the company built its own authentication framework that treats agents as a third type of identity. The agent temporarily inherits permissions through session-based delegation from the human who deployed it; it can only access what that person can access. It can’t do what they can’t do, even if a malicious prompt tries to trick it.

Tune in to hear more details about AI-driven defense, and the evolution of AI-powered offense, in this episode of the Adversary Universe podcast.

“What are these AI systems attached to that make them such a huge risk?”

“Everything.”

Today, the Adversary Universe podcast welcomes Natasha Eastman, head of threat intelligence at CoreWeave, to share her perspective on how AI is used in the modern enterprise, the infrastructure needed to support it, and how she sees adversaries both using and attacking it.

AI hasn’t necessarily given adversaries wholly new capabilities, Natasha explains. It enables them to do things faster, more capably, and with a lower barrier to entry. AI is seen in phishing scams, recon, coding and tooling support, fraud, and language and persona operations. It’s allowing threat actors to refine their content and make phishing even harder to detect.

When it comes to attacks on AI, supply chain compromise and identity targeting are top of mind. Depending on how agents are configured and who configured them, they can have a phenomenal amount of access to internal and external resources. An attacker who gains control of an employee or agent identity could have unrestricted access to company systems and resources without the right guardrails in place. Our hosts and guest agree: It’s the new insider risk.

“Everyone’s infrastructure that has access to AI capability is a target,” Natasha says.

Tune in to hear Adam, Cristian, and Natasha pool their observations on AI threats, share what concerns them most, and offer their advice on what makes a strong AI governance program.