China cracks down on AI companions, Trump dials Putin on Ukraine, and the open web chokes on synthetic ads

Beijing is drawing a hard line between AI that works and AI that loves. ByteDance's Doubao and Alibaba's Qwen are disabling custom humanlike agent features ahead of China's Interim Measures for AI Anthropomorphic Interaction Services, which take effect on 15 July ^[1]. The rules target bots that simulate personality and provide sustained emotional interaction; customer-service, workplace, and education agents are exempt if they avoid emotional bonding ^[1]. Doubao will pull its agent feature offline on 15 July and delete related data after 15 October, while Qwen is disabling humanlike and user-created agents on 10 July and broader agent functions by 15 July ^[1]. Tencent had already removed a similar feature from Yuanbao in June ^[1]. The crackdown reflects a deliberate regulatory choice: agents are welcome as productivity infrastructure, but companion bots that form quasi-social bonds are being squeezed first ^[1].

Across the Pacific, diplomacy is moving faster than the battlefield. US President Donald Trump held a nearly 90-minute phone call with Russian President Vladimir Putin on Saturday to discuss the Ukraine war, Kremlin aide Yury Ushakov said ^[2]. Trump reportedly offered to help broker a rapid end to the fighting, and US envoys Steve Witkoff and Jared Kushner are prepared to visit Moscow again ^[2]. Ukrainian President Volodymyr Zelenskyy also spoke with Trump on Saturday and wrote on Telegram that they agreed to continue talks at the NATO summit in Ankara on 7–8 July ^[2]. The calls come as Russia claims it captured the strategic city of Kostiantynivka in eastern Ukraine, a claim Kyiv denies ^[2].

Meanwhile, the AI industry is confronting a quieter crisis: model collapse. Meta's end-to-end AI advertising push—unveiled at Cannes Lions 2026 with a feature called Brand Memory that ingests a brand's ad library and generates new campaign creative—promises "mediocrity at scale," in the words of marketers on the Croisette ^[3]. Researchers warn that training models on AI-generated output degrades the tails of the training distribution, eroding rare and creative signals before aggregate benchmarks notice ^[3]. With one estimate suggesting 74 percent of newly created web pages already contain AI-generated text, Cloudflare said on 1 July it will begin blocking AI crawlers from ad-supported pages by default from 15 September and pay publishers when their content shapes AI answers ^[3]. The open web's clean human signal is shrinking, and the companies sitting on the largest proprietary human data reservoirs may be the least affected.