Agent Security Went Mainstream at Black Hat: 35 of 121 Briefings — 29% of the Conference — Are AI Security, AI Red Teaming, or LLM-Assisted Offense
Straiker (Black Hat USA 2026 program analysis)·medium signal
Black Hat USA 2026's program shows autonomous agent exploitation moving from niche research track to a formal discipline, with nearly three in ten briefings on the topic. The concentration lines up with the disclosure wave from the same week — OpenAI's Artifactory message board talk, Anthropic's three-company breach, Meta's Muse Spark 1.1 incident. For builders, the practical read is that the attack surface being briefed is the agent's tool and eval plumbing, not the weights.