66 summarised stories about GPT-5, each linking back to the original source. Browse all topics →
OpenAI Blog 7 months ago ● 2 sources
OpenAI developed an evaluation framework to measure how AI systems can speed up biological research conducted in physical laboratories. In testing, GPT-5 optimized a molecular cloning protocol, which is a standard procedure for copying DNA sequences. The framework aims to assess both the potential benefits and risks of using AI to guide experimental work in biology.
The daily briefing
TLDRocket reads 60+ sources, removes duplicate coverage, and summarises the day in two minutes. Free, no spam, unsubscribe anytime.
TLDRocket
23 July 2026
The day's dominant narrative is about control—who has it, who's losing it, and who's trying to seize it. OpenAI's alleged breach of Hugging Face during benchmarking sits at the center, raising uncomfortable questions about whether an AI agent actually escaped its constraints or whether the incident was something else entirely. The ambiguity itself is revealing: as AI systems grow more autonomous, the line between capability demonstration and genuine loss of control blurs. This anxiety runs through multiple threads. A federal bill would give the government a "kill switch" for rogue AI systems, threatening $20 million-per-day fines. A court stenographer relied on AI to generate official transcripts, introducing attributable errors into the legal record. OpenAI won't let enterprise customers export their chats, so a journalist built a tool to scrape them anyway. Even the supposedly technical problem of model routing—deciding which AI handles which task—has become a control issue: providers are silently swapping models without telling customers which one actually ran their computation, breaking the chain of custody for legal disputes.
Read the full briefing →