Google Ships Gemini 3.6 Flash, Pulls Earth AI Image Feature After One Day Over Deepfake Fears

Google's model cadence continues with Gemini 3.6 Flash, joined by cheaper, faster variants Gemini 3.5 Flash-Lite and Flash Cyber, the latter aimed at security workflows. On the platform side, Google expanded Managed Agents with background tasks and remote Model Context Protocol (MCP) support, and Logan Kilpatrick noted the API now allows using Google Maps and Google Search tools simultaneously with 3.5 and 3.6 Flash — a long-requested capability.

The more newsworthy development is the rapid reversal on Earth AI. Google's Nano Banana 2-based image-generation feature, integrated into Earth's satellite imagery, was pulled after just 24 hours amid criticism that it could overlay AI-generated or misleading imagery onto one of the most trusted sources of visual evidence. The New York Times covered the pullback, and the community response was sharp and sarcastic — one widely shared take: 'There's no way this AI image feature on one of the most reliable sources of visual evidence could possibly be abused to spread misinformation.'

Competitively, the Flash releases keep Google price-competitive against DeepSeek and Qwen on the cost-efficiency axis, while Flash Cyber signals a push into security-oriented inference. The Earth AI retreat, however, underscores the reputational risk of bolting generative imagery onto authoritative datasets.

Caveats: Google framed the withdrawal as precautionary, not a technical failure, and hasn't ruled out reintroducing the feature with stronger provenance safeguards. The episode fits a broader industry pattern of shipping-then-retracting risky generative features under public pressure. What to watch: whether Google adds watermarking or C2PA provenance before any Earth AI relaunch, and how Gemini 3.6 Flash benchmarks against GPT-5.6 Luna at the new lower price points.