One-Line Summary

Platforms retreat from risky default AI features while investors highlight fast-moving Chinese model providers to watch.

Big Tech

Meta removes Instagram AI image feature after backlash

Instagram briefly let its AI create new images based on other people’s public posts, and Meta has now removed the feature after heavy criticism. The rollout automatically opted in any account set to public, which meant people’s likenesses could appear in AI-generated images without explicit consent; Meta paused the feature on Jul 10, 2026, saying it “missed the mark” and is “no longer available.” ¹

The feature was part of a wider launch for Muse Image, Meta’s new AI image generator integrated with the Meta AI assistant. On Instagram, special filters created by Muse Image remain available, and Muse Image itself is still accessible on WhatsApp and the stand‑alone Meta AI app. ¹

Hollywood union SAG-AFTRA and privacy advocates criticized the default opt-in approach, calling it a misreading of public sentiment and a misuse of people’s images. Meta said it aimed to provide a creative tool and control over whether public content could be referenced, but acknowledged feedback and removed the feature. ²

Business Insider reports the tool launched on Tuesday and was gone by Friday, with users rapidly sharing opt‑out guidance and talent agencies warning against nonconsensual digital replicas; the episode echoes broader tensions seen with other AI image and video tools. ³

Industry & Biz

Goldman Sachs names preferred Chinese AI model providers

Goldman Sachs analysts identify three favorite Chinese AI model players: Zhipu (Knowledge Atlas), DeepSeek, and ByteDance, and they initiate coverage on Hong Kong–listed Zhipu with a target price of HK$1,880 (about $239.83), roughly 15% above that day’s close. Only Zhipu is publicly traded among the three. ⁴

The report says open‑weight Chinese models are nearing global proprietary performance, with “agentic AI” driving demand at the value end; compute access remains a swing factor. Analysts assessed models on time to market, Arena score, valuation, pricing, and video generation (where ByteDance led); they note Zhipu’s open‑sourced GLM‑5.2 rivals Anthropic’s Fable 5 on several metrics. ⁴

Over the last 60 trading days in Hong Kong, Zhipu shares surged 70% while Minimax fell more than 70%; Alibaba declined nearly 10% and Tencent about 5%. Goldman says Zhipu’s GLM and DeepSeek’s models generally ranked better than Alibaba, Tencent, and Minimax on time to market and Arena score. ⁴

What This Means for You

For marketers, creators, and comms teams, the Instagram reversal is a clear reminder that platform defaults can expose public content to AI use unless you explicitly opt out—carrying brand, legal, and reputational risk. Treat every new AI feature as “on until proven controlled,” then verify settings across brand and personal accounts. ¹

Update policies and contracts to require explicit opt‑in for any AI use of name, image, and likeness, and establish a takedown process for nonconsensual edits. The strong response from SAG‑AFTRA signals where expectations are landing for creators and talent. ²

If you operate in or with China, Goldman’s note implies growing enterprise traction for Zhipu and DeepSeek as cost‑efficient options; ask your vendors which models they embed and how they handle data, pricing, and compute constraints. This helps procurement compare model fit without committing to a build. ⁴

Meta indicates more AI features and integrations are planned across WhatsApp, Facebook, and Messenger, so teams can proactively choose low‑risk tests—like AI filters applied to assets you own—before enabling features that reference public profiles. ²

Action Items

1. Audit Instagram account settings: For all brand and personal handles, review privacy, tagging, and any AI-related toggles; consider private mode on sensitive accounts.
2. Add explicit AI consent to forms: Update influencer, employee, and user‑generated content releases to require opt‑in for AI derivatives of name/image/likeness.
3. Publish a one‑page AI content policy: Define what your team will and won’t upload to AI tools, including escalation paths for takedowns.
4. Ask vendors about model choices in China: If you market or build in China, request which models (e.g., Zhipu, DeepSeek, ByteDance) partners use and their data handling practices.
5. Test safer AI edits first: Try remaining Instagram AI filters or Meta AI edits only on assets you fully own; document results and guardrails for your team.