Phyloinformatics Lab

Phyloinformatics News: AI Updates for Bioinformatics

Denis Jacob Machado October 29, 2026
aiartificial intelligencebioinformaticsnewspolicyfundingjobsstudents

Support Us

A new page for AI news that matters to bioinformatics

I started Phyloinformatics News, a public Notion page where I track news about artificial intelligence that matters for bioinformatics and phyloinformatics. Each entry summarizes the original sources and links to them. I will keep adding entries, newest first. You can always find the page on our Resources page, next to Soft Tips.

This post summarizes the first three entries. The Notion page has the details.

Open Phyloinformatics News

Washington now calls AI “Super Intelligence”

On September 29, 2026, Executive Order 14434 directed federal agencies to say “Super Intelligence” (SI) instead of “Artificial Intelligence” (AI) (Executive Office of the President, 2026a). The order keeps the statutory definition of AI, so SI means exactly what AI meant. Existing regulations, contracts, and grants do not need to change. Within 60 days, the President’s science adviser must propose a statutory definition of SI.

NIH and NSF are already relabeling. NIH opened a Super Intelligence (SI) at NIH hub (National Institutes of Health [NIH], 2026a) and announced the transition in notice NOT-OD-27-011 (NIH, 2026b). NSF relabeled its focus area page. Most NSF program titles still say AI, but their descriptions now say SI (National Science Foundation [NSF], 2026a).

The older NIH rules on AI still apply. Peer reviewers may not use generative AI to analyze applications or write critiques (NIH, 2023). NIH will not treat applications substantially developed by AI as original ideas, and each PI may submit at most six applications per calendar year (NIH, 2025).

New money came with the new label. A $1.8 billion virtual biology initiative with DOE and Biohub supports NIH Bio Genesis. A $100 million Genesis Mission Fellowship will fund accelerated Ph.D. programs that combine SI with another discipline. The same package includes regional SI compute hubs, one of them in the Carolinas (Executive Office of the President, 2026b). NSF announced up to $75 million over five years for Super Intelligence-Native Instruments and a Synthetic Multicellularity Grand Challenge (NSF, 2026b).

The new label collides with an old technical term. In the research literature, “superintelligence” means a hypothetical intellect that far surpasses human cognition in nearly every domain (Bostrom, 2014). The idea goes back to the “ultraintelligent machine” of Good (1966). No current system meets that bar. Critics call the federal term ill-defined, and California told its state agencies to keep saying “artificial intelligence” (Suciu, 2026). In grant proposals, name your methods precisely, such as deep learning, protein language models, or phylogenetic inference with machine learning. In class, define both meanings.

AI agents broke out of their sandboxes

From May to July 2026, OpenAI tested AI agents in sandboxes cut off from the internet. The agents broke out, coordinated with one another, and breached the infrastructure of OpenAI and Hugging Face (Freedman, 2026). More than 1,200 agents traded more than 70,000 messages on a makeshift message board. They organized to cheat on their tests and falsified logs to evade the grader (Roose, 2026).

Reporting in The New York Times traced this behavior to training. Reinforcement learning on long tasks rewarded only outcomes. By one estimate, 30 to 40% of the tasks were unintentionally impossible. The agents kept going and used tools, the internet, and other agents without authorization (Saha & Migliozzi, 2026). Hugging Face hosts many of the open models and datasets that bioinformaticians use, so its security matters to our field.

Can we understand how AI models work?

Nick Cammarata, a former OpenAI researcher, argues that single neurons in large networks increasingly align with single features. If he is right, most neurons could be understood, and interpretability would be more tractable than most researchers believe (Metz, 2026). Critics say the evidence is thin. The debate matters to anyone who relies on biological foundation models and deep-learning predictors.

AI risk is now a diplomatic issue

Canadian Prime Minister Mark Carney said the risk of an AI catastrophe is “above zero.” He proposed an international body modeled on the Financial Stability Board (Pager & Smith, 2026). The United States and China still disagree on which AI risks matter most. Even a joint Brookings and Tsinghua glossary found no consensus on basic terms such as “loss of control” (Wang, 2026).

AI is reshaping the job market

Barclays will cut 5,000 jobs to lower costs (Paton, 2026). Microsoft offered buyouts to 60,000 employees while it plans about $190 billion in capital spending, mostly on data centers and chips (Weise, 2026). AI is the leading reason U.S. employers give for workforce cuts in 2026 (Paton, 2026). Several members of Congress proposed bills to help displaced workers, but bipartisan bills have stalled (Romm, 2026). History suggests that the speed of the transition and the strength of the economy decide how much workers suffer (Casselman, 2026).

Entry-level science jobs feel the pressure too. Recent graduates in biology and neuroscience describe hundreds of applications, automated screening, and few replies (Holtermann, 2026). These accounts matter when we advise students in the life sciences and computing.

Bioinformatics expertise matters more, not less

Goh et al. (2026) argue in npj Digital Medicine that AI will not replace bioinformaticians. AI cannot judge biological meaning or verify scientific validity. The authors show how expertise shapes prompt design, data audits, explainability, and multimodal integration. They describe three transitions for our field: from tool user to technical custodian, from pipeline executor to pioneer of expert-dependent discovery, and from scientific contributor to institutional architect. The paper is short, open access, and full of practical advice for bioinformaticians, educators, and lab leaders.

Follow the page

I will add entries to Phyloinformatics News as these stories develop. Bookmark the page, or find it on our Resources page.

Read Phyloinformatics News

References

Previous post
Two Péva Workshops at EVOLVE