Biosecurity is too important for security theater.

In a quick test with Raygun, we found that DeepMind's SynthIDBio watermark signal could be washed out while preserving the predicted structure.

We don’t think protein sequence watermarks buy much biosecurity.

Researchers familiar with the papers know that this is an unserious use of AI. I work on tax havens so I feel competent enough to go through an example for people from other fields. I chose a published paper because I think including unpublished papers with juniors is evil. 1/8

Excited to share our new @CellStemCell paper led by an amazing PhD student Hamayl! We showed that CAR T, a well-established cancer therapy, remodels muscle “niche”, improves intrinsic regeneration and muscle function in Duchenne muscular dystrophy! cell.com/cell-stem-cell…

Today, in collaboration with Anthropic, we’re excited to share rapid initial progress towards an antibody treatment for food allergy, which affects over 200 million people worldwide, discovered with the help of Claude.

At Divergence Labs, we believe that AI can unlock the next therapeutics with more effective mechanisms than before. Such mechanisms are urgently needed in food allergy, which involve overcoming one of the tightest affinity protein pairs in immunology, IgE and its receptor FcεRI, on mast cells.

Treatment with Xolair—the only approved antibody for food allergy—sequesters circulating IgE, but mast cells remain sensitized for weeks to months due to the extraordinarily slow dissociation of IgE from FcεRI.

We envision a treatment that actively pulls IgE off of FcεRI, without waiting for natural dissociation. Like picking a lock, this requires the right combination of geometry and force.

Our research team, working with Claude, ideated on mechanisms to accomplish this disruption, and converged on an antibody format with two arms with different functions and epitopes. Within one month, we designed, synthesized, and experimentally confirmed the biophysical function of the two arms, which we are now assembling into full therapeutic formats.

@anshulkundaje Opus 4.6 is good (as well as Opus 3, and GPT-5.6 Sol non-Chat)

The Month I Ran My Hedge Fund from the Local Chase Bank

After Enron went bankrupt in 2001, I decided to start a hedge fund. Until then, I’d spent my entire career at a large company, trading off its balance sheet. I knew how to trade, but knew little about the logistics of fund. Plus I had no peers in Houston who could help guide me through the process.

I found out-of-town legal help. My clearing broker was a NY investment bank, Smith Barney. I knew I needed two bank accounts: one for the business to pay rent, salaries, etc and one for the investor money that supported the trades. I figured, incorrectly it turned out, that I could stay local for the latter.

I figured I needed a national, sophisticated bank so I went to my closest Chase branch and asked to open a business account. They were very accommodating and asked some basic questions. Suffice it to say, they were as new to hedge fund finances as I was. But they were happy to have the new accounts, especially since one would be funded with a 6-figure deposit and the other would eventually be funded with millions.

In the spring of 2002, I opened the first account. I used it to pay for normal business expenses: rent, office equipment, legal fees, etc. All was fine.

In August, I called my external investors and told them to send their funds to join my personal investment in the fund to the second Chase account, the limited partnership account. They did. Chase confirmed they received the funds and had no problem.

Quick aside, energy trades in 2002 could be done on an exchange or over-the-counter directly with a counterparty. Most trades could be executed on exchange and be relatively capital efficient but some instruments only traded directly with counterparties. Today, exchanges clear almost any nat gas product one would want to trade but it was less so back then.

I had friends at another trading house, Sempra, who fast-tracked a credit arrangement for me to trade OTC there. I would have to post cash as margin to support my trades given I was not a so-called “credit-worthy counterparty,” but it was fine. I wasn’t planning on taking large net positions given my small capital base, though sometimes I might be long on the exchanges and short to Sempra, or vice versa. I would need to post initial margin and maintenance margin both at my clearing broker and with any direct counterparty to account for the daily price swings at both places. This meant I would have to move money between my prime broker, where I kept my exchange trades, and Sempra, several times a week.

The second week of August I started trading. The first thing I needed to do was send more than half of the LP account funds to my prime broker, where most of my trades would be. The neighborhood Chase branch executed this wire with no problem. Trading at first was small. The second week I did my first trade with Sempra. I had to post margin with them. Chase executed that wire with no problem.

By the third week my trades were picking up. I ended up long with Sempra and short on the exchanges. When the market went up, I had a surplus at Sempra and would ask them to send me those funds, as the credit agreement allowed to me to do. I needed that money to cover the offsetting exchange losses. Very quickly I was moving hundreds of thousands through my local Chase branch several times a week. Red flags quickly went off at the bank. Suffice it to say, I had not prepared them for this level of activity.

The following week the market fell. I needed to send money to Sempra hundreds of thousands of dollars. I told my prime broker to send some of my account surplus there to Chase, and asked Chase to send some of that to Sempra. That’s when my 4 week old hedge fund was hit by an almost fatal blow. Chase accepted the wire from Smith Barney, but I found out the next day that they did not send the wire out. Sempra was calling to ask where the margin money was. I didn’t know. I called Chase and they told me there was a hold on it but it would go out later that day. I checked in the late afternoon. It had not gone out.

I was at the bank went it opened the next morning. I explained how I needed this wire to go out. The account had been frozen, they said, until they got a green light from their compliance department. They didn’t properly “know your customer” and froze my account until they could understand what was going on. This level of activity out of a new account of this size was not something this branch had ever seen before. Nor did they understand an energy hedge fund with over-the-counter trades.

I was furious they had not explained this the previous day and asked them to just send out this one wire and then we’d sort it out. I’d move to another bank asap. They declined. I pleaded, just send this one. It wasn’t working.

I panicked. I had failed to meet my Sempra margin call. They could close my trades. I called Sempra from the bank, explained the issue, and convinced them to give me a 24-hour grace period. But if I didn’t meet my cash call by the next business day, they would close my trades. Then I would be imbalanced on the market with my just my short trades on the exchange. With a sharp market move in the wrong direction, my little hedge fund would have been toast within a month of opening.

Throughout this, normal bank customers were flowing in/out of the bank. I told them I wasn’t leaving the bank branch until they sent the wire. It was my money. I started to raise my voice and cause a scene, the best I could. I hoped they would send the wire just to get me to quit disrupting their normal flow of customers. I’m normally a quiet guy but my business, and maybe career, depended on convincing the branch manager to send this wire out. After two hours and a lot of calls with higher-ups, they consented, but they said they had to send a team to my office the next day to inspect my “business.”

The office space I had was nothing to fancy, to say the least. The building was fine, an 11-story class A. The office, a sublease, was 3 rooms. Open the door and you were immediately on the “trading floor”, which was one row of desks where the 5 of us sat. It had a small kitchen and a tiny conference room to the side. There were a lot of computers but of course no receptionist, no art, no waiting area, just a handful of 20 or 30 something year olds in shorts (August in Houston) with a football nearby.

Chase sent 3 bankers in suits. It was… awkward. I led them to the conference room and tried to explain what we did and why the funds had the velocity they did. They quickly told me that they were not equipped to handle my account and I needed to find another bank by the end of the week. I asked them to facilitate the wires for another couple days until I could, which they agreed to.

I called Smith Barney, my broker in NY, and explained the situation. They immediately understood and helped me open a bank account there. In retrospect, I should have had my investment account there from the beginning. I’d already had my business account at the local Chase and figured they were my bank. I had not properly distinguished between local Chase and JPMorgan Chase. I didn’t know any better.

As I later learned more about banking regs and KYC rules, I realized the local branch acted exactly as to be expected given the circumstance. It was my fault. Luckily, it got straightened out and I can laugh about it now. I still think about how naive I had been every time I drive by the bank branch.

On August 31, the first two participants were injected with @PopVaxIndia’s first fridge-stable RNA vaccine candidate in a Phase I clinical trial in Australia. We built this vaccine at our RNA Foundry in Hyderabad – from computational design to GMP manufacturing. Here's how:

@danintheory proofs are shipping like software now

@davidstutz92alized the model with the watermarked weights and trained it using the multi-step consistency training schedule from Heek et al. (arxiv.org/pdf/2403.06807), with 8 steps. After approximately 4,000 optimizer steps, the logit scores were almost the same as those of the unmarked AlphaFold3 model. The detection rate of my inferred scorer fell to about 1%, compared with nearly 100% at initialization. (This change in logit scores did not occur with consistency distillation i.e., using the teacher model as a guide in Alg 1 of Heek et. al, which is expected.)

I’d be happy to provide our fine-tuned model weights and predicted structures privately to the DeepMind team for independent testing. I’d also welcome confirmation of the scorer and an evaluation using the official detector, so we can check whether our interpretation and findings are accurate. I’ve open-sourced the code for the scorer here github.com/EpiGenomicsCod…. The same repo also contains all the code used for doing discrete multistep CD on boltz-2.

Thanks to @rohitsingh8080’s post on the washing out of the watermark signal with Raygun for inspiring me to dig into this. And also to my collaborators for getting me all the resources I needed to test this out.

My daughter's middle school has the most direct fundraising strategy and I'm here for it.

We can skip the whole charade.

Intriguing thread about stress-testing the structure watermarking scheme of SynthID-Bio. Complements our sequence-based analysis.

I'm not an expert in the techniques here, and will defer to those who know more.

Also, kudos to SynthIDBio folks for making most code/scripts avlbl

Would encourage companies making big announcements to release preprints so that the larger community can provide them real peer review so as to avoid silly claims printed in stone in overpriced glam journals.

And the watermark-scrubbed structures are about as similar to the original (pre-watermark) structures as the DeepMind watermarked structures.

To their credit, the SynthIDBio authors acknowledge that redesign can remove the watermark. But, they argue that doing would still cost time and effort.

Not really. Sequence-to-sequence redesign tools like Raygun mean that sequence watermarks are more theater than security.

This is really vile stuff. No surprise with the title “The Indian Question” (Google JQ if you’re unclear) but this is just a litany of basic xenophobic garbage, with highlights like “white nationalism’s” — meaning a white ethnostate - problem is it’s not strategic. 1/2

I sent this message to my student, whose two papers received scores of 5/4/4 and 5/4/3 at #NeurIPS2026 but were rejected.

It’s also what I want to say to all junior students going through the same process: please be proud of what you accomplished with these papers, and don’t let our confidence in research.

Happy Mid-Autumn Festival to you all! 🌕 🥮

Bio problem statements that read like bad zen koans, trading actual scientific depth for profound-sounding headlines.

You know, #RECOMB 2027 is in Toronto next year. Submit your papers and posters, and eat around probably the best food city in North America.

nytimes.com/2026/09/18/tra…