Editorial briefs · Page 9
Not the fastest feed. A useful judgement built from several signals.
DeepSeek Harness just went viral, then faced 14,560 controlled attacks
A recent paper ran 14,560 controlled indirect prompt-injection tests against DeepSeek Harness; one hidden-Unicode file attack combination reached a 25.5% rule-judge success rate, using local fixtures without real external side effects.
For agent, MCP, and terminal-tool users tracking prompt-injection risk
DeepSeek opened an all-weekend off-peak lane less than a week after changing prices
Multiple outlets report that DeepSeek API will charge Saturday and Sunday entirely at off-peak rates from August 23, while weekday peak windows remain. Reported V4 Pro output rates are RMB 27 at peak and RMB 13.5 off-peak per million tokens. The public pricing page has not separately listed the weekend exception, so model and account scope still require invoice-level verification.
For DeepSeek API users running batch jobs or long-running agents and watching costs
What is happening at OpenAI? Twelve executives left in a year, and the org chart may be the busiest product
Business Insider listed 12 prominent OpenAI executives and leaders who left in 2026 for reasons including startups, health, role changes, and governance disagreements. It is not proof of one feud, but it shows the organization is repeatedly redrawing its boundaries.
For readers tracking OpenAI products, leadership changes, and AI-company governance
Why is ChatGPT swearing more lately? Users say it is becoming less like a customer-service bot
Business Insider reported that multiple users noticed more profanity from ChatGPT around the switch to GPT-5.6 Luna for free users; OpenAI has not confirmed an intentional product change.
For everyday ChatGPT users tracking personalization and tone changes
Three days after the SpaceX deal, a Cursor employee used $200 Ultra access to lure Claude users
At least one user publicly said a Claude cancellation screenshot earned roughly $200 in Cursor Ultra access. It was not cash or a confirmed long-term promotion, but it exposes a new phase in AI coding competition: subsidizing migration and fighting for workflows.
For Claude Code, Cursor, Codex, and Grok users tracking the fight for developer workflows
Claude's watermark launches and a 14.8k-star project fires back
The watermarks-remover surge shows resistance to equating AI involvement with AI authorship, but nobody can honestly prove Claude's mark is cleared until Anthropic's detector is available.
For Claude users and anyone tracking AI authorship, provenance, and compliance
The important thing about Qwen 3.8 27B is not who it beats
A 27B open model is now part of local deployment, benchmark, and agent conversations. The real shift is that owning the runtime is becoming practical again.
For people who want local AI without relying on marketing leaderboards
ChatGPT for teens is not just another mode: safety is becoming product surface
OpenAI's focus here is not a smarter model, but age controls, parental controls, healthy-use features, and learning as a default experience.
For parents, teachers, and anyone watching the boundaries of AI products
Coding AI is leaving the chat window: the race is to finish and ship
Codex case studies, Claude Code workspace features, and OpenClaw fixes point to one shift: the value of a coding agent is closing the loop in a real environment, not writing a snippet.
For developers, product people, and anyone applying AI to real projects
ChatGPT ads reach Europe: answers are starting to become distribution
OpenAI says ChatGPT Ads is expanding to 31 European markets. It does not replace search overnight, but it makes AI comparison and decision moments a new commercial entry point.
For people working on brands, content, websites, and product distribution