2026 · 06 · 26 · Fri

Updates for 6/26

Gemini Flash now has screen control built in; Anthropic accused Alibaba of 28.8M distillation calls. Adobe bought Topaz; ChatGPT long-paste went free.

A · Theme of the day

Computer-use AI drops into the Flash price tier

Gemini 3.5 Flash now handles screen-control natively, no extra harness needed.

Gemini Flash bakes screen control in

Gemini (Google)Gemini (Google)
Compared to before

Six months ago, AI that could operate a browser meant flagship Pro pricing plus a complex agent-harness setup.

What changed

Gemini 3.5 Flash now has computer-control built into the model — click, type, scroll — no separate agent harness needed.

Why it matters

At Flash pricing, browser-automation PoCs become much more realistic. For personal hobby projects, the difference is negligible.

B · Theme of the day

Gen-AI turf reshaping: free unlock, acquisition, and IP war

Free-tier unlock, acquisition, and distillation lawsuit all in one day.

ChatGPT long-paste lands on free users

GPT (OpenAI)GPT (OpenAI)
Compared to before

Until last month, only Plus/Pro users had this protection; free-tier pastes expanded into the full conversation and could be unintentionally memorized.

What changed

Texts over 10,000 chars now attach as separate files instead of flooding the chat context — rolling out to Free and Go plans.

Why it matters

A quiet but real win for daily free users who paste in long docs or logs. Rarely matters if you paste shorter than a paragraph.

Adobe buys Topaz Labs, owns the enhancement stack

Adobe FireflyAdobe Firefly
Compared to before

Firefly covered generation, but refining existing footage at high quality still meant reaching for a third-party tool like Topaz.

What changed

Adobe acquires Topaz Labs (super-resolution, denoising, frame-interpolation) — next: integrating into Photoshop, Premiere Pro, and Lightroom.

Why it matters

Video post-pro teams already using Topaz will likely move into Adobe's ecosystem. Irrelevant for 3D and audio workflows.

Anthropic names Alibaba in 28.8M distillation attack

Claude (Anthropic)Claude (Anthropic)
Compared to before

Anthropic's ToS has always banned training rival models on Claude's output, but this is the first time they publicly named a specific company in violation.

What changed

Alibaba-linked accounts queried the Claude API 28.8M+ times allegedly to harvest training data for competing models — in violation of Anthropic's ToS.

Why it matters

The US-China AI race just moved into IP-lawsuit territory. Enterprises using AI APIs should review their own ToS compliance now.

Archive

Past updates

A daily archive of changes actually applied to the site.