AI news this month reaches well beyond new model releases. Researchers are testing whether agents can help make scientific discoveries, new data is showing where people use AI at work, and policymakers are debating how to oversee increasingly capable systems. Here are three developments reported through September 27, 2026, with links to the original announcements.
September 23: Claude helps identify a previously uncharacterized enzyme system
Anthropic reported early results from its new life sciences lab. Claude agents searched a large DNA database and noticed repeated sequences near a reverse transcriptase gene. After further analysis and laboratory testing, Anthropic’s scientists described a previously uncharacterized system they call array-associated reverse transcriptases (ARTs).
The arrangement resembles features of systems such as CRISPR, but the primary function of ARTs remains unknown. This is not yet a new gene-editing tool or a medical treatment. The result, shared in a preprint linked from Anthropic’s announcement, is an example of AI helping scientists find promising leads that still require experimental follow-up.
September 15: Google maps how people use AI at work and in science
Google expanded its AI & Economy ATLAS, an open tool for exploring AI usage across occupations and countries. In the data Google presented, arts, design, and media account for 19% of work-related AI usage in India. Computer and mathematical occupations account for 30% of work-related AI usage in the United States. These figures describe shares of observed work-related AI usage, not the percentage of workers using AI.
Google also highlighted a survey of more than 600 scientists in the US and UK. Nearly half of respondents said they use some form of AI daily, and they reported saving just under seven hours a week. The researchers caution that time saved does not automatically produce discoveries: AI outputs need validation, while physical experiments and clinical testing can become the next bottlenecks.
September 9: OpenAI calls for mandatory AI safety requirements
In a policy post, OpenAI backed mandatory national safety requirements in the United States based on model capabilities. It also said it supports four California bills addressing independent safety assessments, auditor standards, protections for young people, and biological risks.
This is OpenAI’s policy position, not a set of rules already in force. For teams building AI products, it shows how evaluation, auditing, and safeguards are becoming central to the public discussion around deployment.
What connects these stories?
AI is moving into more varied, concrete work. In the lab, it can help surface candidates for scientists to test. In workplaces, usage data is beginning to show meaningful differences across fields and countries. As capabilities grow, questions about oversight grow with them. All three stories are still developing: scientific findings need further validation, adoption figures need careful interpretation, and proposed rules must pass through the legislative process.