Kyiv Under Fire on Eve of NATO Summit as Google and Tencent Shake Up AI

Overnight Russian strikes killed at least 23 people in the Kyiv region and left residential blocks partially collapsed, Ukrainian officials said, in the second large-scale assault on the capital in less than a week ^[1]. The Ukrainian Air Force warned that a "serious shortage" of interceptor missiles meant none of the 23 ballistic missiles fired were shot down, though cruise missiles and drones were intercepted ^[1]. President Volodymyr Zelensky is now pressing allies for "strong decisions" on air defense at this week's NATO summit in Ankara, where he is expected to meet U.S. President Donald Trump on July 8 ^[1][2].

The timing is critical. Kyiv hopes to use the summit to secure more Patriot systems and interceptors, and to discuss strategy for pressuring Moscow into direct peace talks ^[2]. Trump, who recently told Zelensky to act "more boldly," is expected to present his own vision for a peace agreement and to speak with Vladimir Putin afterward ^[2].

In tech, Google's quiet June privacy update is drawing fresh scrutiny. The company now says media uploaded to Search services—including images sent through Google Lens, audio from voice search, and recordings from Google Translate—can be saved and used to "develop and improve Google services and technologies, including AI models" ^[3]. The new "Search Services History" and "Personalized Recommendations" settings are on by default, meaning users must actively uncheck the "Save Media" box to opt out ^[3].

Meanwhile, Tencent's Hunyuan team released the full version of its Hy3 model under the permissive Apache 2.0 license, a shift from an earlier preview that had restricted commercial use ^[4]. The 295-billion-parameter Mixture-of-Experts model is roughly half the size of Zhipu AI's GLM-5.2 and trails it on coding benchmarks, but Tencent claims Hy3 leads open-weight models on agentic search, tool orchestration, and long-context retrieval ^[4]. The company also reported cutting the model's hallucination rate from 12.5% to 5.4% in internal evaluations ^[4].

The agent-evaluation space is heating up too. San Francisco-based Patronus AI raised $50 million in Series B funding, bringing its total to $70 million, to stress-test AI agents in simulated digital environments ^[5]. Founded by former Meta AI researchers, the startup builds synthetic worlds that replicate websites and internal systems so developers can catch failure modes before deployment ^[5].

Together, the day's headlines show AI's growing footprint in both consumer privacy debates and military-diplomatic strategy—two arenas where the stakes are rising fast.