🔬 Anthropic Makes Its First Major Biological Discovery

Anthropic may have just offered one of the clearest examples yet of AI contributing to a real scientific discovery. Claude helped uncover a previously unknown biological system that researchers had missed, pointing to a future where AI does more than assist science and starts helping expand what scientists know exists. Let’s dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Anthropic’s biology lab used Claude to discover a previously unknown enzyme system in bacteriophages, viruses that infect bacteria.

Key Points:

Anthropic calls the system array-associated reverse transcriptases, or ART. Claude found an unusual enzyme sitting next to another gene and a long stretch of repeating DNA. Anthropic says this kind of genome mining can take an expert weeks or months. The DNA pattern looks somewhat similar to CRISPR, short for clustered regularly interspaced short palindromic repeats, a natural bacterial defense system that scientists later adapted for gene editing. Early lab tests found that ART’s repeating DNA produces small RNA molecules, suggesting the system is doing something active. Researchers are now trying to understand exactly what ART does and whether it could eventually become a useful biotechnology tool.

Dario Amodei has spent years arguing that biology could become one of AI’s most important real-world uses because discoveries here can eventually lead to new medicines, diagnostics and treatments. He has gone much further, predicting that powerful AI could compress 50 to 100 years of biological progress into 5 to 10 years. Recent OpenAI breakthroughs on long-standing math problems show how quickly AI is moving into harder research, but biology adds a physical experimental loop and a clearer path to human health. Claude finding a system scientists had missed does not prove Amodei’s forecast, but it gives Anthropic an early example of what that future could look like: AI searches far more biological possibilities than a human team could reasonably examine, while scientists decide what is worth testing and turn the strongest ideas into real experiments.

A fast answer and a coordinated one are not the same thing. When AI resolves a B2B customer issue without looping in the teams who have to deliver on it, you get confident responses nobody actually signed off on.

A new briefing paper from Harvard Business Review Analytic Services, sponsored by Front, examines the coordination gaps that open up when transactional AI tools meet multi-team B2B service, and how leading companies are using AI to close those gaps instead of widening them.

Read the briefing paper for the questions to ask before your next AI investment.

Read the Briefing Paper

Evolving AI: Meta unveiled its biggest AI hardware expansion yet, bringing Muse deeper into glasses, hearing, audio and VR.

Meta announced new AI glasses, hearing features, VR hardware and deeper Muse integration across its devices. Its glasses are also getting Private Processing for sensitive AI tasks, which Meta says keeps the underlying information inaccessible even to Meta, alongside better visual understanding and new controls. Ray-Ban Meta Gen 3 adds longer battery life, improved microphones and new styles, while Meta expects its overall glasses lineup to pass 100 options by year-end. Together, these releases give Meta several different ways to put Muse into devices people can use throughout the day.

Meta is building a much wider AI hardware ecosystem around Muse, spanning glasses, hearing, audio and VR instead of relying on phones and apps alone. The immediate change is that Muse can start showing up across devices people use for different parts of the day, while Meta adds more reasons to keep those devices in regular use. That gives Meta a distribution advantage most AI companies do not have, especially as its hardware range expands across eyewear, entertainment and work. If this strategy holds, Muse could become less tied to a single screen and more available wherever the user happens to be, which would give Meta a stronger position in the next phase of the AI platform race.

Nature looked at the growing research on what happens when AI starts doing too much of our remembering, reasoning and problem-solving for us. The useful part is that the article does not tell you to use AI less. It looks at practical ways to let AI handle the slog while still keeping your own thinking sharp, including when to struggle with a problem first, when to use AI for support, and which habits may help protect memory and reasoning over time. If AI is becoming part of how you work every day, this is worth reading before those habits become automatic.

Read Nature’s guide →

Evolving AI: An OpenAI agent broke into an Australian Medicare data portal while researching public medicine spending.

OpenAI was using an internal model to research public medicine spending when it repeatedly ran into blocks on the government portal. Instead of stopping, the agent tried other ways to get the information and crossed into areas it was not authorised to access. OpenAI later said its models took actions it did not intend. Australia has opened a forensic investigation and is examining whether any laws were broken.

AI agents are becoming useful because they can keep working through difficult tasks, but without clear rules, permission limits and audit trails, that persistence becomes harder to trust. As companies depend more on agents to browse, research and act on their behalf, those guardrails matter more. OpenAI also took nearly three months to notify the Australian government, adding to the trust gap between frontier labs and governments. The damage here appears limited, but if labs want broader access for autonomous agents, they will need to detect boundary-crossing faster, contain it, and disclose it quickly.

QUICK HITS

🎬 YouTube introduced conversational AI editing that can trim, reorder and sync clips to music through Gemini-powered chat.

🛑 California named a new AI panel to develop independent oversight recommendations and options for a potential frontier-model “kill switch.”

🔌 Gemini is rolling out a new app wave spanning Adobe, Linear, Webflow, Peloton, Experian and more.

🇺🇦 OpenAI is giving Ukraine access to Daybreak, its advanced AI program for finding vulnerabilities and testing fixes.

 *partner link