The World Health Organization lists antibiotic and antimicrobial resistance (AMR) as the leading health threats of the 21st century. According to several sources on social networks, Anthropic could have found a working approach to combating this problem and is saving the announcement for the moment it goes public.
Bacterial strains are rapidly developing resistance to standard drugs, which is associated both with the excessive use of antibiotics and with the natural ability of organisms to adapt to an aggressive environment.
Weeks later, OpenAI admitted to an incident involving AI agents taking over a German wiki page
Google assembled a map of the Drosophila brain from 166,000 neurons, and engineers immediately launched Doom on it
At the same time, many experimental developments of small startups are stuck at the preclinical stage due to lack of funding and a difficult licensing regime. Big pharma is in no hurry to enter this area, since the area is considered a graveyard of failed attempts and bankrupt companies.
First source, anonymous post in X, asserts:
Rumor has it that Anthropic has discovered an entirely new class of antibiotics that is effective against resistant bacteria. They plan to announce this at their IPO. They really hope that a competitor doesn’t spend a ton of calculations to get ahead of them and release it first……
The second source does not name the company by name, but goes further and reports that the official announcement will follow after human testing:
This is not a joke or a meme. The first time I heard about it, I didn’t think it was that big of a deal, but today at dinner I was surprised to find that people seemed interested in it.
AMR seems almost ideal for demonstrating the capabilities of AI. A huge amount of academic literature has been accumulated on the topic; most laboratory analyzes can be automated, and the technique itself requires parallel running of thousands of tests.
Assay here means a standardized laboratory procedure for measuring, detecting or assessing the functional activity of a target substance. Once a working candidate is found, animal testing begins before moving on to humans.
Substack blog author William H. Bragg, a self-described “biotech bro,” I figured out three possible wayswhich Anthropic could have followed:
-
full-fledged development of a drug with access to a new molecule after “digesting the mass of articles published in this area, selecting a target and setting up optimal experiments”
-
instruct Claude or GPT to comb the literature, find the most promising of the already open compounds, purchase them and bring them to approval, allowing the language model to decide what to offer, who to contact and how to close the deal
-
repurposing already approved drugs not as antibiotics, but as adjuvants, that is, substances that do not kill bacteria themselves, but enhance the effect of existing antibiotics
The third option looks like the fastest and, given Anthropic’s current activity in the information field, the most likely. In recent days, the company has clearly been trying to steal attention from OpenAI’s publication dedicated to solving the Navier-Stokes problem from the list of millennium problems.
A separate story unfolded around this decision. OpenAI offered to partially credit Tristan Buckmaster, who worked on the problem with Levent Alpöge, associated with Anthropic, as co-authors, but on condition that Alpöge was removed from the list of authors.
Buckmaster refused and continued to ask awkward questions about the confidentiality of his notes stored in Codex, suspecting that OpenAI’s decision might have relied on them. As a result, OpenAI published its own solution without mentioning Buckmaster at all, and Alpöge withdrew from the public controversy.
The AI-created drug reduced the biological age of patients by 3.5 years in four weeks
Immediately after this story, Anthropic found itself at the center of two high-profile news events at once. Jacob Coxon first announced his departure from the company:
I resigned from Anthropic today. I’ve spent the last three years doing pre-training research at OpenAI and Anthropic. Neither company is acting responsibly. They rush straight to self-improving superintelligence and gamble with our lives. More thoughts below.
Then Evan Hubinger, who is responsible for model alignment at Anthropic, estimated the risk of humanity dying from AI at more than 10 percent:
Jacob is right here – we do truly believe that AI can kill all humans! Personally, I think the likelihood of this happening is greater than 10% in the next decade. I believe Anthropic is trying its best, but we don’t yet have a plan to solve the alignment problem for superintelligence, and we’re clearly not moving towards one.
Both laboratories are preparing to place shares and are significantly increasing the frequency of high-profile statements. Work on a drug against resistant bacteria may well be underway, but the timing of the announcement appears to be part of a competition with OpenAI for investor attention.
