When the Attacker Already Knows the Building
This week's episode is different. No headlines, no CVEs — just Bishop Fox Red Teamers: Brandon Kovacs, Leron Gray, Thomas Wilson, and Rob Antonucci, talking through what it actually looks like to operate at the intersection of AI, custom tooling, physical access, and adversary emulation.
AI is a force multiplier only if you already know what you're doing. The team was blunt: red teamers who aren't using AI to accelerate tool development are falling behind — not just in speed, but in the quality of coverage they're giving clients. The qualifier matters though. The value isn't in generating code blindly; it's in having a requirements document in your head and using AI to close the gap between knowing what you need and having the dev time to build it. Handcrafted artisanal code, as Leron put it, still beats slop — but now you can iterate toward it faster. The team also flagged a practical edge: vendor detection signatures are largely public, and AI knows them. You can now build tooling that avoids known patterns by design, not by luck.
Off-the-shelf tools get caught. Custom payloads don't. The era of downloading certify.exe and running it is over for real engagements. Rob walked through a project where the team had a clear domain admin path and couldn't execute it — because nothing they had would get past SentinelOne. That failure pushed Rob toward years of payload development work. The lesson the team drew wasn't defensive; it was about what that ceiling means for attackers: the floor for getting caught is rising, which means the operators worth worrying about are the ones who've already solved this problem before they show up.
Chaining network and physical is what a real nation-state looks like. The most technically impressive story from the session was an engagement for a high-profile energy-sector client where the team ran a full assume-breach-to-physical chain. They found a file share with home-encrypted credentials — key stored next to the ciphertext — cracked them, got to domain admin, pivoted to the physical access control servers, cracked the application hashes, and created fake employee records with photos, names, and door access. Brandon then walked into the building with a badge they'd minted, with an operator 3,000 miles away unlocking doors in real time from the compromised ACS. The real unlock: the SOC re-enabled the one payload S1 caught because the filename looked like a legitimate SQL DLL. Human review, human error, engagement survives.
Social engineering scales because humans haven't changed. AI now removes the language barrier from phishing entirely — not just translation, but naturalistic register, which is the thing that used to give foreign-origin attacks away. Thomas delivered a stack of pizzas to a New York skyscraper, bribed the lobby guard with a slice, and photographed the ethernet ports and desk layout while the internal contact went to get him a pen. The social engineering thread throughout the episode wasn't about tricks — it was about the consistency of human behavior under social pressure. People let the pizza guy in. People re-enable flagged payloads when the filename looks right. The technical controls aren't where the story ends.
Critical infrastructure is on the public internet and almost nobody has tested it. The team closed on a thread that ran through several stories: satellites accessible via public URLs, gas station fuel systems with no authentication, train control dashboards reachable from the open internet. Rob mentioned a recent engagement where they accessed a system that controlled currently-orbiting satellites — and the client's response was essentially: great finding, please stop immediately. Thomas pointed to the gap that these systems have been running for decades on protocols that predate modern security thinking — and most have never seen a pen tester.
The takeaway. The biggest breaches don't necessarily come from zero days. They come from the pizza guy, the re-enabled payload, and the satellite that's been on the internet since 2009.
Subscribe to our PODCAST
Real talk on the threats, trends, and tactics shaping security today
Recommened Resources
This site uses cookies to provide you with a great user experience. By continuing to use our website, you consent to the use of cookies. To find out more about the cookies we use, please see our Privacy Policy.