This playbook lays out the intersections between cybersecurity and machine learning systems, with an explicit focus on LLM threats, adversarial approaches, and the rise of agentic capabilities. It frames those topics as an ecosystem issue rather than a single-technology problem, and signals that conversations at Black Hat USA 2025 and DEF CON 33 will foreground both offensive and defensive dynamics. The article identifies core themes: how large language models change attacker tradecraft, how adversarial techniques can be used against and inside models, and why red‑teaming at scale has become a practical requirement for teams shipping model-driven products.
The analysis is oriented to startup teams. It positions startups as both potential innovators and vulnerable targets, particularly when they integrate third-party models or build agentic features. The playbook does not offer prescriptive recipes in this preview; instead it promises a strategic lens that maps technical exposures to enterprise decisions, product roadmaps, and threat modeling. Key terms in the title—agentic Artificial Intelligence, LLM threats, adversarial AI, AI sovereignty, and red‑teaming at scale—are presented as interconnected vectors that influence risk, compliance, and market positioning for early-stage companies.
Access to the full playbook is subscriber-gated. The article page describes subscriber benefits including exclusive research, a weekly OODA Network Dispatch, and a community Slack workspace for practitioners and experts. Metadata on the piece lists Daniel Pereira as the author and shows a publication date of 08/08/2025. Pereira is identified as research director at OODA with more than 20 years of experience in foresight strategy, creative technology, and ICT research; the preview frames the full report as intended to inform startup teams planning attendance at security conferences and to prepare them for the strategic and operational conversations that will dominate those events.
