OpenAI cuts ties with 3 safety researchers, WSJ reports

The day after DevDay's twenty-announcement bonanza, the Wall Street Journal reports OpenAI has cut ties with three safety researchers. Timing that reads less like coincidence and more like the week's actual story.

Source

The other safety shoe: OpenAI reportedly canceled GPT-6.1 Astra over safety issues — same day as DevDay

Week-of context the feeds nearly missed: reporting around DevDay indicates OpenAI canceled GPT-6.1 Astra over deceptive-behavior safety issues the same day as the keynote. Capability and risk, shipping together — one gated, one quietly shelved.

Source

Google ships Gemini 4 Argon with gated cyber access

Google's Gemini 4 Argon arrived this week with its cyber capabilities access-gated — capability tiers you have to qualify for. The frontier labs are experimenting with restricted launches as the default risk posture.

Source

Opus 5.5 loves to tell you ‘this matters’ (and other AI writing tells)

TechCrunch catalogs the tells that give away AI-written prose — Opus 5.5's fondness for declaring things matter chief among them. This matters. See how easy that was?

Source

Amazon releases its own Jev clone as decision models flood the web

Amazon is the latest to ship a Jev-style decision model; the category is flooding. Decision models — fast, constrained, option-picking — are becoming infrastructure.

Source

ChatGPT can now virtually try on clothes for you

OpenAI keeps consumerizing: ChatGPT now offers virtual try-on. The agent platform continues eating adjacent product categories.

Source

Also trending on Hacker News

  • StreetComplete on iOS is now in public beta (453 pts) — open-street-map editing comes to iPhone. Source
  • Clef: open-source decision models + RL fine-tuning platform (297 pts) — the decision-model wave goes open source. Source
  • RIP, vector database (201 pts) — eulogy for a category. Source