AI News: July 5, 2026

1. Alibaba Reportedly Bans Employees From Using Claude Code

Alibaba. Alibaba will bar employees from using Anthropic’s Claude Code starting July 10, 2026, after classifying it as high-risk software over security concerns, according to reports. The move follows disclosures that Anthropic embedded code to identify Chinese users, which an Anthropic engineer described as a March experiment for account protection, and Alibaba is steering staff toward its own Qoder coding tool instead. The episode shows how geopolitics and data-access rules are fragmenting the market for AI coding assistants. Source

2. Midjourney Seeks to Force Hollywood Studios to Disclose Their Own AI Use

Midjourney. Midjourney is trying to compel Disney, Universal, and Warner Bros. to reveal their internal AI usage in an ongoing copyright suit in which the studios accuse the company’s models of unlawfully generating images of their characters. Midjourney argues the studios are cherry-picking documents that support their market-harm claims while concealing their own AI development practices. The discovery fight could shape how much of a plaintiff’s own AI activity becomes relevant in generative-AI copyright cases. Source

3. Study of 26,000 Students Finds AI’s Learning Cost Surfaces Over Two Years

Researchers. A study by Strömberg and colleagues tracked more than 26,000 secondary students in central China over 30 months and found that AI use led to faster homework and higher assignment grades but a roughly 20% drop in closed-book exam scores within six months. The full damage to high-stakes entrance exams took about two years to appear, with declines of 18 to 24%, and around 81% of long-term AI users showed patterns of outsourcing work rather than learning with the tool. The authors warn that short-term studies miss the long-term learning cost of AI assistance. Source

4. Greg Brockman Envisions an ‘Almost No Interface’ Future for Software

OpenAI. OpenAI cofounder Greg Brockman described a future of “almost no interface,” where AI assistants handle digital tasks autonomously and people no longer need to learn software at all. He framed ChatGPT less as a feature-packed app and more as an invisible layer that acts on a user’s behalf. The vision underscores how agentic systems could reshape software design and the skills users are expected to bring. Source

5. Open-Source Tool pxpipe Encodes Text in PNGs to Cut Agent Token Costs

pxpipe. A new open-source tool called pxpipe encodes text into PNG images so coding agents such as Claude Code and Fable 5 process it as visual input, reportedly cutting token costs by up to 70%. The approach targets the rising expense of feeding large contexts to agents that bill per token. It is an early example of developers engineering workarounds to the per-token pricing of agentic coding tools. Source