Announced, Not Yet Shown: Four Fields Where the AI Claim Outruns the Evidence
This week's four items have one thing in common: in each, the AI is described by the people who built it, and the evidence for it is thinner than the announcement. A genome-language-model company says its software designed a phage for one patient. A trapped-ion company posts a record error rate and mentions AI decoders in passing. A fusion company gets federal money to build an AI co-pilot for a machine that is not yet running. And a brain implant's first speech result comes with no description of its decoder at all. We note below where each claim stops.
Radical Numerics Says Its Genome Model Designed a Phage Given to One Patient
On October 20, 2026, Radical Numerics issued a press release launching Omnii-1, which it describes as "a biological AI model that learns across DNA, RNA, proteins and other molecular data." The company says the model converts 28 data types into a common representation and is being shown on eight applications, including personalized cancer vaccine design and rare-disease genetics. The headline claim is that researchers using Omnii designed a bacteriophage, a virus that kills bacteria, which was given to a patient at an unnamed clinical site under physician compassionate-use access. In lab testing, a cocktail containing the AI-designed phage killed the patient's bacterial isolate before administration. The company says the design took weeks rather than the months a phage search often takes.
Weigh it accordingly. This is a press release, not a paper: it names no treating physicians, gives no site, and reports no patient outcome. It is one patient and one bacterial isolate, and compassionate use is not a trial. The outside physician quoted speaks about how hard phage searches are in general; the release does not say she treated this patient. This is a genome-scale design story rather than gene editing in the strict sense, and nothing here has been independently checked.
Quantinuum Reports 99.9995% Logical Fidelity on Helios, With a Color Code
On October 19, 2026, Quantinuum announced a logical fidelity of 99.9995%, which it describes as just five errors in a million operations, on its trapped-ion Helios computer using a color code. The company says that is 10 times lower error than its own September 2026 record, set with a different code called Helix, and 100 times lower than the best published result on other hardware, which it attributes to a Google Quantum AI superconducting system. It says each increase in code distance cut remaining logical errors by a factor of 12, and that the color code used up to 37 physical qubits per logical qubit. This ran on real hardware, not a simulation. Robin Blume-Kohout of Sandia National Laboratories, who was not involved, is quoted calling the paper "real, serious performance gains."
The AI angle is thin. The release says the architecture "integrates tailored AI decoders" and mentions real-time decoding, but it does not say what kind of decoder was used in these runs, or how much of the result depends on it. The comparison figures are the company's own, and the results were achieved in experimental settings.
Commonwealth Fusion Will Lead a 10-Partner AI Digital Twin of SPARC
On October 8, 2026, Commonwealth Fusion Systems (CFS) announced it had been selected for a Phase II award under the U.S. Department of Energy's Genesis Mission, to lead an AI-enabled digital twin of SPARC. Partners are Lawrence Livermore, Oak Ridge, Princeton Plasma Physics Laboratory and SLAC national laboratories; Carnegie Mellon University and MIT; and Google DeepMind, NVIDIA, Synopsys and General Atomics. The listed components include training datasets and AI surrogate models, reinforcement learning and agentic models for an AI-based SPARC operations co-pilot, and visualization pipelines. CFS principal investigator Tom Looby is quoted saying AI models trained on DOE supercomputers "will be used to plan pulses in the SPARC control room," turning "hours of heavy simulations into instant insights between plasma pulses." The Fusion Report, a trade newsletter, relayed the news on October 16 and put the Phase II awards at $159 million across 12 projects; CFS did not disclose its own amount.
This is a plan, not a result. SPARC is still being assembled in Devens, Massachusetts, the project is multiyear, and the announcement gives no milestone dates. CFS is the party making every claim here.
Paradromics Reports Its First Long-Term Participant Generating Speech in Real Time
On September 14, 2026, CBS Austin reported that a Michigan woman with a progressive motor neuron disease is the first participant in Paradromics' FDA-approved Connect-One early feasibility study. Neurosurgeon Dr. Matthew Willsey and his team at the University of Michigan implanted the Connexus device in June. The company says the system decoded her first freely chosen sentence, "Okay, I have a lot to say," in real time with no errors, and that she used it during a live phone call with her grandchildren through synthesized voice and text. The article says software converts signals from hundreds of recording channels into text and speech.
This is six weeks old and the weakest AI item in the issue: the article does not describe the decoding software as AI or say how it works, and a second trade report gave no vocabulary size, accuracy or latency. "No errors" is Paradromics' own claim about one participant. The device is investigational, and the company plans to follow her for six years to study safety and durability.
The Thread
The capability curve this site follows shows up differently in each item: a model that claims to design, a decoder named but not described, a co-pilot not yet piloting anything, and software that turns spikes into sentences without saying how. All four are announcements by the builders, and none has outside confirmation yet. That is itself the pattern worth noting: much of what the field says about AI is ahead of what it has shown.
Quantinuum's own release title describes its work as validating a path to fault-tolerant quantum computing by 2029 Forecast. It is a company target, not a result, and the Helios numbers above are the evidence offered for it.
Sources
- Omnii-1 / AI-designed phage: Radical Numerics press release, October 20, 2026
- Helios color-code result: Quantinuum press release, October 19, 2026
- SPARC digital twin: Commonwealth Fusion Systems, October 8, 2026; The Fusion Report, October 16, 2026
- Paradromics Connexus: CBS Austin, September 14, 2026
- Nov 22026 No. 9Doubled in Cells, Simulated in Silicon, Promised in a Post: Four Fields, Four Distances from Proof A protein language model designed a DNA-insertion enzyme that doubled targeted insertion in cultured cells. NVIDIA simulated 104 fault-tolerant qubits with no AI in the description. Commerce signed a letter of intent for up to $85 million for Commonwealth Fusion electronics. And Musk says Neuralink's Blindsight is engineering-ready, with no data.
- Oct 262026 No. 8Announced, Not Yet Shown: Four Fields Where the AI Claim Outruns the Evidence Radical Numerics says its genome model designed a phage for one patient, Quantinuum posts a 99.9995% logical fidelity with AI decoders mentioned in passing, Commonwealth Fusion will lead a 10-partner AI digital twin of an unbuilt SPARC, and Paradromics reports first real-time speech from one participant. Every claim is the builder's own.
- Oct 192026 No. 6The Puddle Problem: What the Weak Anthropic Principle Might Say About Our Epoch of Firsts A puddle thinks the hole was made for it. The weak anthropic principle explains why: observers can only find themselves where observers can exist. This issue works through the puddle, WWII bombers, and the age of the universe, then asks how much of Issue 3's "epoch of firsts" is just a puddle effect, and what survives.
- Oct 122026 No. 5Decoders and Drift: Four Fields Learn to Read a Moving Signal Penn used machine learning to steer a compact gene editor to 97% efficiency at its best target, and a neural decoder claims a 17x lower logical error rate on a quantum code, in simulation. Jefferson Lab modeled magnet drift at DIII-D, and UCSF decoded speech and gesture from one implant.
- Oct 52026 No. 4Models in the Loop: Four Fields, Four Levels of Proof Infleqtion claims 30 logical qubits with a gate found by an AI model, though its code only detects errors. PPPL's PACMAN let machine-learning models act on a real tokamak. Neuralink showed a third ALS participant speaking through a decoder. And AI-designed gene editors are still only proven in cells.
- Sep 282026 No. 3The Window: Why We May Be Living in the Moment of Firsts Many of the biggest events in human history are happening for the first time, within a few generations: controlled nuclear energy, a complete human genome, the first gene-editing drug, error-corrected quantum computing, galaxies seen 290 million years after the Big Bang, and a deflected asteroid. This issue lays out the evidence, then asks what it implies for the L factor in the Drake equation, and whether a window this narrow means a civilization this short-lived.
- Sep 212026 No. 2Sequences, Simulations, and a Wired Avatar Anthropic says Claude searched bacterial DNA and surfaced a CRISPR-like enzyme system of unknown function, and outside scientists are cautious. IonQ showed a real-time error-correction decoder handling 408 simulated logical qubits on one CPU. Commonwealth Fusion's SPARC ran its cryogenics at full 8 K operating temperature. And UCSF decoded speech and gestures at once from two paralyzed patients to drive a full-body avatar.
- Sep 142026 No. 1Four Frontiers, One Curve Beam Therapeutics has now dosed 25+ patients with an AI-refined base editor for hereditary ATTR amyloidosis. Rigetti's Ankaa-3 hit 84 qubits at ~99% fidelity while Google's AlphaQubit keeps out-decoding classical error correction. Commonwealth Fusion's SPARC crossed 75% construction with its first 20-tesla magnet installed. And Eon Systems loaded a real fly-brain connectome into physics simulation and watched it walk, untrained. Four different fields, one shared engine showing up inside each of them.