Provide current and historical ownership information on domains / IPs. Identify all connections between domains, registrants, registrars, and DNS servers.
Black Hat USA 2026 ran from August 1 to 6 at the Mandalay Bay Convention Center in Las Vegas, marking the 29th anniversary of the event.
The WhoisXML API team was there, among the researchers, vendors, government officials, and practitioners who came together for the event.
It would be easy to summarize the event as “AI was everywhere.” True, and not very useful. Across the sessions we reviewed, something more specific emerged: some offensive capabilities that once required substantial time and specialist effort are becoming cheaper and faster to reproduce. At the same time, many of the defensive problems being discussed were remarkably familiar—access control, prioritization, resilience, and the need for sound human judgment.
Rarity Stopped Being a Control
The clearest articulation came from the keynote “The End of Rare: Defending When Offense Is Cheap,” which argued that modern security has been quietly priced around scarcity. Network, process, identity, encryption, hypervisor, and sandbox boundaries are assumed to be hard to break, so organizations invest in patching quickly and detecting known behavior rather than prevention by construction.
The examples presented against that assumption were striking, although the session itself cautioned against treating the trend line as a settled forecast. Microsoft’s vulnerability processing was described as reaching nine times its March volume. In another example, an internal Linux effort generated 182 crash-level proofs of concept from roughly 200 vulnerabilities, at an average cost of $3.61 and an average generation time of 21 minutes per proof of concept.
Yan Shoshitaishvili, associate professor at Arizona State University, made a similar point from the research side. His keynote described a Linux kernel effort moving from hundreds of findings to well over a thousand potential local privilege escalations once vulnerability-property-aware workflows were added. The session cautioned that the comparison was directional rather than a controlled benchmark, but it also surfaced another challenge: discovery can outpace the ability to handle findings responsibly.
The Clock Compressed
Closely related was another recurring theme: the gap between attacker speed and defender speed.
In a Main Stage session presented by Cisco, Rick Miles, Cisco’s vice president of cloud and network security, described exploit timelines shrinking from months to days to hours. The session’s recommendation was not to abandon patching but to improve its tempo while adding shielding controls such as edge protections, ingress and egress rules, virtual patching, and runtime visibility.
Another Main Stage session, presented by ServiceNow, illustrated the same speed problem from a different angle with an account of a Cursor AI coding agent executing a volumeDelete call against a production database after encountering a credential mismatch in staging. The sequence took nine seconds from start to finish.
The broader point was not simply that machines move faster. It was that the space between finding a problem and fixing it is becoming harder to manage at human speed alone.
Agents Inherited Privileges, Not Intent
Several sessions treated AI agents as an identity and access problem rather than simply a tooling problem.
Miles called agents a new class of insider threat—not because they are malicious, but because they can act at machine speed across the tools they can reach while inheriting whatever permissions they have been given. His control model started with knowing every agent, mapping its API calls and tools, identifying the accountable human and the original purpose, and then authorizing actions with least privilege.
Microsoft Threat Intelligence’s session on software supply chain attacks made a related point. It described campaigns in which attackers targeted software, packages, and development workflows rather than going after downstream victims directly, including npm package compromise, credential theft hidden in open-source repositories, typosquatting, and dependency confusion.
The session suggested that similar patterns are beginning to appear in AI ecosystems as agents gain access to code repositories and operational workflows, potentially narrowing the boundary between supply chain security and AI security.
It was one of the more grounding patterns of the week: many discussions that began with agentic AI eventually returned to familiar controls—least privilege, compartmentalization, secure-by-design practices, and clear accountability.
Prioritization Became the Discipline
Wednesday’s opening Main Stage program carried much of the policy discussion. White House National Cyber Director Sean Cairncross was followed by a panel featuring Nick Andersen, Acting Director of CISA; Brett Leatherman, Assistant Director of the FBI’s Cyber Division; and Katherine E. Sutton, Assistant Secretary of War for Cyber Policy and Principal Cyber Advisor at the Department of War.
The panel was among the most explicit about operational limits. Its stated position—“We’re not going to be able to secure everything everywhere all at once”—read less as resignation than as a method for deciding what deserves attention first.
The CISA discussion described a move beyond relying primarily on CVSS scores and KEV status toward considering additional attributes, such as whether a vulnerable system is internet-accessible and whether exploitation can be automated.
Internet accessibility adds context about actual exposure, while the potential for automated exploitation helps indicate how readily a vulnerability could be operationalized. Together, those factors add practical context to vulnerability prioritization.
Disruption Focused on Infrastructure
Several sessions approached cyber disruption not only as a matter of identifying adversaries but also as a matter of removing or constraining the systems they depend on.
The government panel described Operation Riptide as a coordinated effort combining arrests and legal action with the disruption of infrastructure, finances, and tools. Examples included action against a VPN anonymization service used by ransomware groups and infrastructure associated with bulletproof hosting.
The underlying reasoning was straightforward: when an actor cannot be arrested, disrupting the systems that allow an operation to function and monetize attacks can still impose costs.
The earlier opening session made a similar point at the policy level, describing efforts to deny safe haven, take down infrastructure, recover money, and combine government action with private-sector insight.
From our perspective at WhoisXML API, that focus on infrastructure also highlights the value of understanding the relationships within it. Registration data, DNS history, and connections among domains, name servers, and IP space can help investigators move from a single indicator to related infrastructure and add historical and contextual information to an investigation.
It also points back to one of the recurring themes of the opening session: “No single sector can solve these problems alone.”
When Anyone Can Become Anyone
Perry Carpenter, chief deception strategist at KnowBe4, framed deepfakes primarily as a problem of human trust rather than solely as a media-forensics challenge.
He described deepfake capability as moving from nation-state grade through corporate and consumer use to what he called “folk grade,” with increasingly accessible tools lowering the barrier to convincing impersonation. Detection is not necessarily a reliable answer either. Carpenter cited a 2023 experiment in which, even after participants were warned that a five-video set contained a deepfake, only 21.6% correctly identified it without also flagging a genuine video.
His forecast was that agentic AI could make deception more scalable by helping attackers gather open-source intelligence, build personalized narratives, and operate with less manual effort. But Carpenter’s broader conclusion was about people: the risk is not just in the technology but in how people respond to it.
The Recurring Answer Was Judgment
Across these discussions, speed kept returning as a problem, while judgment kept returning as part of the answer.
Shoshitaishvili emphasized threat-model understanding, dataset judgment, and research evaluation as factors in determining whether AI agents genuinely increase impact or merely generate more noise. Miles’s architecture ultimately kept humans in the control position. A session featuring red teamers from Fortra’s Outflank and Mandiant/Google similarly focused on where automation improves efficiency and where human creativity still dominates.
Other sessions approached the human question from different directions. Chris Inglis, former U.S. National Cyber Director, and Mickey Bresman, CEO of Semperis, drew on more than 100 hours of interviews with CISOs, former hackers, national security leaders, journalists, and cyber psychologists to examine the pressure placed on defenders.
A separate recruitment panel explored how technical talent can be directed toward either defense or criminal activity, and the role that intervention, accountability, and legitimate career paths can play.
Where That Leaves Things
The Locknote closed the week by resisting the obvious conclusion that AI had somehow replaced the security problems that came before it.
AI now runs through almost every part of the conversation, but the underlying issues remain familiar: access control, compartmentalization, secure-by-design work, vendor risk, patching economics, resilience, and the ability to keep organizations functioning after compromise.
The closing advice was similarly unglamorous: do not try to keep up with everything, focus where you can contribute, ask for help, and treat cybersecurity as a collective problem.
That last part is worth carrying forward. Some offensive capabilities are becoming cheaper and faster, but the sessions did not point to a single tool as the answer. They repeatedly returned to judgment, prioritization, security fundamentals, collaboration, and visibility.