AI Bioweapon Warning and Bezos Betting Big on Brain Science

Sam Altman and Dario Amodei agreeing on something publicly is, by itself, newsworthy. The fact that they've co-signed the same open letter to Congress — alongside Microsoft, Meta, and Google DeepMind — tells you just how seriously the AI industry is taking the threat of biological weapons.

The letter, organized by two Washington think tanks, asks Congress to make one specific thing mandatory: screening synthetic DNA and RNA orders for dangerous genetic sequences. Right now, the biggest suppliers of lab-ready genetic material do screen purchases, but they do it voluntarily. No law requires it. No federal agency enforces it. And in a world where AI can help someone design a novel pathogen and then order the genetic building blocks online, that gap is starting to look less like an oversight and more like a liability.

Here's the part that should make you sit up straight. Synthesizing dangerous pathogens used to require deep expertise, expensive equipment, and access to a serious research facility. That combination of barriers kept the worst-case scenarios largely theoretical. But AI tools are systematically dismantling those barriers — making it easier to design sequences, understand what makes them dangerous, and figure out how to assemble them. The expertise floor is dropping fast.

Among the signatories is Demis Hassabis of Google DeepMind, who won the 2024 Nobel Prize in Chemistry for AI-driven protein prediction work. That's not a coincidence. The same class of AI that earned a Nobel Prize for predicting how proteins fold is also theoretically capable of helping someone engineer proteins that shouldn't exist outside a containment lab.

The letter doesn't call for slowing down AI development or restricting synthetic biology broadly. The ask is narrower and more surgical: require screening at the point of sale, keep detailed records of orders, and build a paper trail that gives authorities a shot at catching threats that slip through initial checks. It's closer to a background check system than a ban.

What makes this moment unusual is the coalition behind it. The AI industry fights about almost everything — model safety, open source, regulation, compute access. But the signatories here span direct competitors, nonprofit researchers, national security veterans, and the biotech companies that actually sell synthetic genetic material. Twist Bioscience and Ansa Biotechnologies, both major players in the synthetic DNA market, signed on too. When the suppliers are asking to be regulated, it's worth paying attention.

Congress has historically moved slowly on tech-adjacent biosecurity issues, partly because the science is hard to explain and partly because the threat feels abstract until it isn't. The letter's authors seem aware of that dynamic, which is probably why they leaned into the rarity of the moment. Getting Altman, Amodei, Suleyman, and Hassabis to agree on anything is a feat. Getting them to lobby Congress together is something else entirely.

Jeff Bezos wrote a $50 million check for a company that, at the time of pitching, had no product, no proven technology, and no clear answer to the central question it was trying to solve. That question: how does the human brain actually learn?

The startup is called Flourish, and its founding bet is that modern AI is architecturally broken in a way that can only be fixed by going back to biology. Rob Williams, a former Amazon S-team executive who helped build Alexa, and Thomas Reardon, a neuroscientist and serial founder, are the people making that bet. They pitched Bezos the Amazon way — a mock press release written as if the product already existed — and he apparently liked what he read.

The core argument is one that AI researchers have whispered about for years but rarely built companies around. Large language models consume staggering amounts of energy and data, and once training ends, they stop learning entirely. A single AI chip burns more than 30 times the power a human brain uses to process information. Training a frontier model requires ingesting essentially everything humanity has ever written, and the next model needs even more. It's a scaling treadmill with an increasingly steep energy bill.

The human brain, by contrast, runs on roughly 20 watts — less than a standard light bulb — and a child learns a language from a few hundred thousand spoken examples, not billions of text tokens. Reardon's framing is blunt: there is something fundamentally wrong with a system that needs to read every book ever written dozens of times just to learn English. A toddler does it faster, cheaper, and without a data center.

Flourishwants to build what they're calling Cortex AI: a synthetic intelligence system that matches the brain's computational efficiency, learns continuously after deployment, and runs on 50 watts or less. It's an audacious target, and the founders are the first to admit they don't yet know how to hit it.

What they do have is a plan to find out. The company is assembling a team of AI researchers and neuroscientists working in parallel, running original wet lab experiments with high-end equipment to extract usable architectural insights from actual brain tissue. The idea is that the neuroscience informs the engineering, and the engineering asks better questions of the neuroscience. Whether that loop produces a breakthrough or an expensive dead end remains genuinely unknown.

Bezos wasn't the only one willing to take that risk. The total funding round reached $500 million, with additional backing from Lux Capital among others. That's a serious amount of money to spend hunting for something that may or may not exist in a form that's reproducible in silicon.

But the upside case is enormous. If Flourish or anyone else cracks continuous, efficient learning at low power, it would make the current generation of AI look like a very expensive prototype. The brain has been solving this problem for millions of years. The question is whether a well-funded team can reverse-engineer the answer in time to matter.