🩸 The US military wants fake blood. The fake war wins were already free.

The US military just put $73 million behind blood that can ride in a backpack, an Osaka chip read amino-acid handedness one molecule at a time, three lab-grown mini-brains teamed up to sort Braille, a Khosla-backed startup began freezing people’s immune cells for $4,995, and Claude sketched scorpion-toxin binders that stick but don’t cure. Five stories, and every headline sprinted ahead of its methods section.

NEWS

The US military has had a loud year of declaring victory, so it is worth noting the one win it is actually buying: blood that survives a backpack. The earlier wins were the ones where the Pentagon declared victory in the Iran war while the fact-checkers argued. DARPA is putting $73 million into a program called RAPIID, aiming to field freeze-dried plasma, platelets, and red-cell substitutes by 2029. technologyreview Hemorrhage drives about 90% of potentially survivable battlefield deaths, per Brian Eastridge’s 2012 Joint Trauma System review in JAMA, and a 2017 JAMA study of combat casualties in Afghanistan tied prehospital transfusion to a 74% lower risk of death over 24 hours, with the median transfusion starting 36 minutes after injury.

Two of the products show why anyone bothers. KaloCyte’s ErythroMer packs hemoglobin salvaged from expiring donor blood into a lipid nanoparticle, freeze-dries it, and keeps it stable at room temperature for two years; in rodents that lost up to 70% of their blood, it held blood pressure without the vessel-clamping that sank earlier products. Case Western’s SynthoPlate, a synthetic platelet, cut bleeding and blood loss by more than half in mice with low platelet counts. DARPA frames RAPIID as a 36-month push toward a fully FDA-authorized, fieldable system, and in October it kept signing performers, including up to $10.7 million for RxMP Therapeutics. prnewswire

The release skips the graveyard. HemAssist, a 1990s hemoglobin carrier, saw 24 of the 52 people who got it in a trial die, versus 8 of 46 controls, and a 2008 JAMA meta-analysis of 16 trials and 3,711 patients jamanetwork found the class “associated with a significantly increased risk of death and MI,” a relative death risk of 1.30 and a nearly threefold higher chance of heart attack. Shibani Pati of UCSF told that same MIT Technology Review feature that backing products untested in humans can make things harder, and warned against starting with “moonshot projects that will take forever to get through.”

The blood is real and the need is real. The 2029 deadline is a wish. The fake wins came cheaper.

RESEARCH

Life on Earth builds its proteins almost entirely from left-handed amino acids, so a lopsided left-right ratio is one of the tidier chemical hints that something alive was involved. A team led by the University of Osaka built a chip that reads that handedness one molecule at a time, and ScienceAlert framed it as a chip that could spot signs of alien life. Science Alert

The device is two gold electrodes separated by a gap about half a nanometer wide. When a single amino acid slips into the gap, it changes the quantum tunneling current, and a machine-learning model sorts the resulting squiggle. In a paper in Nature Communications, the team reports distinguishing L- and D-forms with over 80% accuracy, physosaka-u which lead author Takahito Ohshiro and senior author Masateru Taniguchi call the first discrimination of amino-acid chirality at the single-molecule level. osaka-uosaka-u They then ran extracts from the Murchison meteorite and Atacama Desert soil through it.

The “passed its first test” headline skips a step. Telling identity and handedness apart across all 39 forms scored an F1 of 0.49, better than the 0.026 of random guessing but not a clean read. thedebrief The classifier was trained only on known amino acids, so on the real extracts it rejected most events as unknown and kept only a fraction, and its D/L ratios disagreed with standard mass-spec on several amino acids. The paper, received in 2023 and accepted in September, calls this a targeted, high-confidence method rather than comprehensive profiling.

It is also not a long-range detector. You still have to physically hand it a sample, which for now means Mars has to come to the chip. NASA helped pay for the work, so at least someone with a budget is looking in the right direction.

RESEARCH

The headline is that lab-grown mini-brains read Braille. The quieter truth is that they twitched, and an ordinary classifier did the reading. Still a neat result, and a cousin to the rat-neuron video model we covered last month, though this one is a peer-reviewed paper built on three-dimensional organoids rather than a product launch.

A University of Bristol team grew human forebrain organoids, each a half-millimeter blob of about 10,000 neurons, on FinalSpark’s remotely accessible Neuroplatform, which rents organoid time at $1,000 a month. They converted a tactile sensor’s scans of the 26 Braille letters into electrical pulses, zapped the organoids, and recorded which electrodes fired.

Those firing patterns went into a support vector machine, a simple machine-learning algorithm that did the actual sorting. frontiersinfrontiersin One organoid’s output decoded to the right letter about 61% of the time; pool three organoids and it climbed to 83%, the figure in the paper’s abstract in Frontiers in Artificial Intelligence. arxivarxiv Under added Gaussian noise the three-organoid ensemble held that 83% while decoding straight from the stimulation pattern fell to 55%.

A few caveats keep it honest. This is open-loop: the organoids received signals and responded, and the scientists stress the experiments do not show whether the organoids can learn over time. techxplore The organoids are not computing so much as adding biological texture that a classifier then exploits, which is reservoir computing’s whole trick. It was posted as a preprint in August and is now peer-reviewed, and the three Bristol authors, none of them FinalSpark staff, declare no competing interest even though FinalSpark supplied the organoids.

Three brains beat one, nobody is sure why, and the Braille stayed on the page the whole time.

NEWS

A startup called Reservoir would like to freeze your immune cells today so some future medicine can use a younger version of you. The pitch, from co-founder Nabiha Saklayen’s launch post, is blunt: “Billionaires have owned their cells for years. Starting today, you can too.” twitter

So what are you buying? A phlebotomist visits your home, draws about 60 milliliters of blood, and Reservoir isolates the immune cells and stores them below minus 150 degrees Celsius in an FDA-registered facility for $4,995 up front, then $270 a year. withreservoir It is built by Cellino, the cell-manufacturing company Saklayen runs, launched in partnership with Vinod Khosla, and backed by Khosla Ventures, 8VC, and Leaps by Bayer, with Reservoir billed as a preservation service rather than a treatment. financialcontent

Now the fine print. The science people reach for is future cell therapy, like CAR-T, which today needs far more cells than one blood draw yields. Saklayen told Longevity Technology the company banks five vials of about five million cells each, a deliberate trade-off against an hours-long apheresis, and does not screen your cells or claim any therapeutic use. longevity One of its own partner clinicians, Jay Luthar, called much of what is marketed in this space “unproven, or even has the potential to cause harm,” and framed the offering as “optionality.” longevity Optionality with an annual fee.

The honest comparison is private cord-blood banking, which sold the same biological-insurance promise and whose banked units, by most accounts, sit unused. Scientific American Reservoir is betting that medicine will someday want your 2026 cells. It might. You are paying storage while it decides.

RESEARCH & NEWS

In September an Anthropic model found an enzyme nobody could explain. This time it designed proteins that grab scorpion toxins, and the lab numbers are cleaner. An internal Claude model, steered by the investment research firm Citrini Research, generated 318,484 designs over 64 hours, and two labs, Adaptyv Bio and Twist Bioscience, tested 211 of them, of which 17 bound a scorpion toxin. anthropic

The targets were five neurotoxins from four scorpion species. Scorpions sting more than 1.2 million people a year and kill at least 3,250, most of them children, nih so the problem is worth the compute. The 17 hits split across four toxins, with the two strongest latching onto Cn2, a Mexican-scorpion toxin, in the single-digit nanomolar range, and competing for the same spot as an antibody known to neutralize it in mice. anthropic None bound more than one toxin, which was the harder and more useful prize.

Anthropic says so itself: “we do not claim to have produced an antivenom.” anthropic The binders stuck to toxins grown from genes rather than real venom. Nothing was tested in cells or animals, and binding is not the same as neutralizing. Outside experts agreed, with one telling The Verge that finding a binder is “easy-ish” and “the hard part is to translate it into a medicine”. mezha

Worth naming: Citrini is an investment shop, and after the experiment it turned the result into a stock thesis favoring drug-testing firms like Charles River. xqz A tidy number from people who stand to trade on it is still worth reading twice. The binding is real. The cure is not here yet.

Notice how much of this edition lives in the future tense. The blood ships in 2029, if the trials hold. The chip finds alien life once someone hands it a sample. The organoids compute once a classifier does the sorting. Your frozen cells matter once medicine learns what to do with them. The binders become antivenom once something finally stops a toxin in an animal. Every result is real. The promises are the part still waiting on a sequel nobody has greenlit.

The exception is the scorpion count: 1.2 million stings and more than 3,250 deaths a year, mostly kids, happening in the present tense while the pipeline takes its time.

Which future-tense promise are you actually betting on? Hit reply, we read everything. And if you know someone who keeps a chest freezer they call an investment, forward this along.

Keep questioning everything (especially anything sold in the future tense),

Prateek & Jere

P.S. In September Claude found an enzyme nobody could explain. Last month, binders nobody has cured a mouse with. Two for two on “technically real, not yet useful.”