Safety · Featured
It arrives looking like a safety certificate, but it is closer to a brochure the manufacturer chose to print — and the most revealing part is often the section that isn’t there.
Consumer ChatGPT uses your chats to train OpenAI’s models unless you opt out — a switch that lives in two places, doesn’t reach data already used, and which some users report finding turned back on.
This week Anthropic stopped serving minors, Google’s Gemini app asked to train on you before it would open, and OpenAI’s limits moved again. A week of AI quietly raising the price of admission — in ID, data and patience.
Studies from NYU, Stanford and Veracode keep finding that 40 to 45% of AI-generated code ships a known vulnerability — and that the assistant’s confidence makes developers less likely to check. Here is the mechanism, and what to do about it.
From 14 September, DeepSeek routes every V4 Pro API call to the cheaper V4.1 Flash, with no Pro-tier option until an undated V4.1 Pro. It is a real price cut wrapped around a model swap you did not choose.
OpenAI's GPT-6 Astra launched to headlines on 3 September, but paying ChatGPT users were largely locked out, Sam Altman apologised for a 'messy' rollout, and Europe got no data-residency option at all — a launch that outran its own availability.
Google's Gemini app quietly tightened its filters — users say it now refuses benign creative writing and even a plain Markdown question.
No on-site comments — no backend, no accounts, nothing tracked. Corrections, counter-arguments and war stories are welcome on GitHub Discussions.
One email, occasionally, when something is worth complaining about. No spam, because that would be hypocritical.
No spam. Unsubscribe anytime. We never sell your address. Opens the confirmation in a new tab.
Prefer to chip in? This blog is reader-supported.
