Marcus Thompson specializes in cybersecurity, model security, and extreme risks related to AI. Formerly a security engineer at a major cloud provider, he later joined an AI red-teaming group where he attacked language models, vision systems, and data pipelines. He approaches every new AI technology with a penetration tester's instinct: where it breaks, how it can be exploited, and what the real attack surfaces look like.

Marcus writes about prompt injection, data leakage in RAG systems, supply-chain security for models, and vulnerabilities tied to GPUs and compute infrastructure. He dissects security reports, tests marketing claims against real practices, and evaluates the robustness of mitigation strategies. His tone is dry, fact-driven, and focused on concrete abuse scenarios rather than abstract speculation.

His longer features reach into geopolitics: industrial espionage, export controls on advanced chips, state-level competition for compute capacity, and the role of criminal organizations in exploiting AI systems. At AI-Telegraph, Marcus heads the "Security & Risk" section, designed for CISOs, architects, regulators, and engineers who must take decisions in environments defined by adaptive adversaries and rapidly shifting technology.