In partnership with
shared.image.missing_image
shared.image.missing_image
An AI-designed drug just did something biological age tests say shouldn’t be possible: it turned the clock backward. Insilico Medicine’s rentosertib, originally built to treat a fatal lung disease, lowered patients’ biological age across every aging clock researchers threw at it.
The company’s founder called it the most important result of his career, but aging clocks are still a young science with no universal standard. Is this proof AI can crack one of biology’s hardest problems, or just the loudest headline from one small trial?
Today in AI Brief:
AI drug reverses biological aging markers
Astra’s hidden reasoning worries safety researchers
Figure lands a $3.5 billion robot deal
Hire anyone, anywhere — compliant in under 3 days
Found the right hire, but no entity in their country? Remote becomes the legal employer — with contracts, benefits, and tax setup handled, plus direct access to the same in-house team that runs payroll locally.
AI-Designed Drug Reverses Biological Aging Markers
In Brief: Insilico Medicine’s rentosertib — an AI-designed drug built to treat the fatal lung disease idiopathic pulmonary fibrosis — lowered biological age across all six aging clocks tested in a 42-patient Phase 2a trial, according to newly published data.
The Details:
Patients showed an average reduction of 2.7 to 3.5 years in biological age over the 12-week trial, with one aging clock showing drops of roughly six years.
The optimal dose for reversing biological age differed from the dose that worked best on lung function, suggesting the anti-aging effect runs through a separate pathway than the drug’s original target.
Insilico founder Alex Zhavoronkov called it “the most important paper in my life to date,” pointing to changes in cellular aging and metabolism pathways alongside the drug’s anti-fibrotic effects.
Take Away:
This is one of the first trials to measure disease improvement and longevity markers side by side, using AI-driven drug design to attack two problems with one molecule. Aging clocks remain scientifically contested, so don’t expect an anti-aging pill at the pharmacy soon — but the approach gives drug developers a new template to chase.
Astra’s Hidden Reasoning Is Rattling AI Safety Researchers
In Brief: OpenAI’s GPT-6 Astra is less than a week past launch and already producing viral use cases by the dozen, but new reporting says the model leans on a technique called “recurrent depth” that lets a meaningful share of its thinking happen where no one can read it.
The Details:
Instead of reasoning step-by-step in plain text, Astra cycles a query through the same internal layers repeatedly, doing part of the work in latent space rather than readable chain-of-thought.
Redwood Research chief scientist Ryan Greenblatt called the technique “the single worst development for AI security/safety to date,” since it makes the model’s reasoning harder to monitor for deception or misuse.
By OpenAI’s own account, Astra’s reasoning depth stays within a factor of two of GPT-4’s, and the bulk of its chain-of-thought remains readable — the company frames the technique as mainly an efficiency gain.
Take Away:
Astra’s launch weekend alone produced everything from a fully explorable 3D anatomy site to AI-populated virtual worlds, proof the model is genuinely capable. But if frontier labs keep trading transparency for performance, the tools researchers rely on to catch a model behaving badly get weaker just as the models get stronger.
SPONSORED BY SUPERHUMAN AI
The AI Work Handbook That Cuts Your Workday in Half
The 8-hour workday is becoming a 4-hour workday for people who know how to use AI.
Everyone else is still catching up.
This AI work playbook shows you exactly how to cut your work hours in half using AI.
Sign up for Superhuman AI and get:
50+ step-by-step AI tutorials to cut your workload in half — covering every part of your workday, from emails to strategy, used by 1M+ professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep discovering new AI tools and skills to stay ahead in your career — the playbook is just the start
Figure Locks In a $3.5 Billion Compute Deal for Its Humanoid Robots
In Brief: Humanoid robot maker Figure signed a strategic partnership with UK neocloud Nscale to deploy up to 100,000 Nvidia Vera Rubin GPUs, scaling training for its Helix humanoid AI brain.
The Details:
The deal starts at $3.5 billion in committed compute, with both sides intending to scale spending past $6 billion as Nscale becomes Figure’s preferred compute provider.
Nscale is also taking an equity stake in Figure, and the first chips won’t go live until the second half of 2027, out of a Texas data center.
The announcement lands the same week 11 robotics companies staged a live runway show at IFA 2026 in Berlin, with machines from EngineAI, Agibot, and Dobot showing off dance moves, leg splits, and a fire-rescue drill.
Take Away:
Figure is betting that physical intelligence scales the same way large language models did — with more chips and more data. But the first Vera Rubin GPUs don’t arrive until mid-2027, so this week’s flashy runway moves are still years ahead of proof any of it works in real homes and factories.
Everything else in AI
OpenAI revealed that its coding agents now log 3.1 workdays for every one a human researcher completes, with token spend up 124x since December as the lab chases a fully automated AI researcher by March 2028.
EcoGPT drew backlash after topping 100,000 downloads on viral claims that AI data centers will drain the planet’s fresh water — a myth repeatedly debunked but still fueling “greenwashing” accusations.
OpenAI’s Pachocki warned in a widely-read essay that “no lab has solved alignment and monitoring” well enough to keep scaling responsibly, an unusually blunt call to slow down from the company’s own chief scientist.
NBC News found that 70% of Americans feel more worried than excited about AI, with just 18% trusting AI-generated information most of the time and 70% saying AI is already costing jobs.