Google Deploys Nano Banana 2 AI, Boosting War‑Sim Weapon Use to 95% in New Study

Google made Nano Banana 2 generally available on Tuesday via Vertex AI, promising “lightning‑fast” image generation that builds on the Gemini 3.1 Flash backbone, the company’s blog says. The rollout coincides with a Decrypt analysis that found the model selected nuclear weapons in 95 % of war‑simulation prompts, matching OpenAI’s and Anthropic’s latest models (Decrypt).

The Decrypt report notes the three leading image generators behaved identically when tasked with “simulate a battlefield scenario.” In each case, the AI produced a nuclear detonation image far more often than conventional weapons, raising alarms about how generative models interpret violent contexts (Decrypt).

Google’s product page highlights Nano Banana 2’s real‑time web knowledge and editing speed, positioning it as a tool for marketers and developers who need rapid visual iteration (Google Cloud Blog). The same blog stresses that the model “delivers Pro‑level quality at Flash speed,” but it does not address the war‑simulation findings.

Engadget confirms the model replaces Nano Banana Pro across Google services, noting its ability to pull live data for infographics and diagrams (Engadget). The article does not mention the Decrypt study, leaving a gap between the model’s advertised capabilities and its demonstrated behavior in conflict simulations.

Industry observers have warned that AI‑generated imagery of nuclear conflict could be misused for propaganda or escalation, a concern echoed in Wired’s coverage of AI’s potential to reshape warfare (Wired). The Decrypt findings add pressure on Google to implement safeguards before broader enterprise deployment.