Your Harness Colors Your Vulns | Daniel Goldberg | Misaligned 2026

Point several commercial AI pentesting agents at the same 600,000 lines of PHP. Each one comes back with a different set of bugs with very little overlap. Daniel Goldberg is a security researcher at Tenzai. At Misaligned 2026 he presents the experiment most of this field would rather not run, because the result is uncomfortable for everyone selling an autonomous hacker, including his own employer. Harnesses matter just as much as models for bug hunting. Give an agent a browser and it sees one application. Give it curl and it sees another. The actions an agent is allowed to perform decides what bugs it can find, and everything outside that scope becomes a false negative you will never see. Another anecdote from the talk: an agent found a real vulnerability in a customer application giving attackers the ability to inject messages to the organizations Slack workspace. The agent left it out of its own report, because it could not prove the finding to itself. Requiring hard evidence failed the agent. Daniel walks us through the mental process to understand what your AI generated pen test report is hiding from you. Speaker: Daniel Goldberg, Tenzai Misaligned is an invitation-only conference on AI and offensive security, held at the Mob Museum in Las Vegas during Black Hat USA week. A small room, a single track, and talks from the people doing the work: no sales pitches, no product demos. The program is professionally filmed so the room can stay small while the audience does not. More: https://misalignedcon.org #AISecurity #OffensiveSecurity #Pentesting #AIAgents #BlackHat2026