12/22/25, 1:10 PM

Darktrace’s year-ahead piece, published on 22 December 2025, argues that 2026 will hinge on how organisations secure AI models, agents and the identities powering them, with AI and automation driving both risk and opportunity. It flags agentic AI as a potential insider risk, noting that agents can be prompted into unintended sharing of data or misrouted communications if access and behaviour are not tightly controlled.

The analysis highlights prompt injection as a leading breach vector, predicting indirect prompts could expose data or cause agents to act outside their intended scope. It also asserts that humans will be outpaced by machine-speed operations, urging defenses to assume human fallibility and rely on automated provenance checks and dual-channel verification.

Additional trends include SaaS platforms becoming prime supply‑chain targets, and the continued commercialisation of generative AI and AI assistants in cybercrime, which could scale attack frameworks. Taken together, the article contends that securing AI will be inseparable from securing the business, with organisations needing to understand how AI is used, behaves and can be misused to prepare for 2026.