Block

Block

The Washington Post reports that Senate Republicans have unveiled a spending bill that would block the White House from implementing the rule change proposed by the Office of Management and Budget that would give political appointees sway over federal research grants (we covered this proposed rule change last month in The Brief).

The provision is part of a funding bill that will keep the government funded through December 11 of this year.  From the Post’s story:

OMB said it planned to finalize the rule by Oct. 1. If the Senate’s funding extension were approved by both chambers, it would block OMB from implementing the rule, at least until the measure expires on Dec. 11.

The measure effectively punts a more permanent block until after the midterm elections. Most polling and pundits expect at least one chamber of the next Congress to be controlled by Democrats—meaning the likelihood of a more permanent block is high.

The move by Senate Republicans may have been influenced by the especially strong pushback that the proposed rule change received from across the scientific world. 496,769 comments were submitted in response to the proposed rule change. As Stat News reports, an analysis of 52,322 of the comments that were posted in full indicates that 95% of the comments are opposed to the proposal and only 1% support it.

Meanwhile, Science reports that as much as $1 billion in National Science Foundation funding earmarked for basic research is reportedly being diverted to a new initiative called X-Labs meant to transfer NSF-funded research to industry. While the concept of X-Labs is not controversial (many universities have technology transfer offices), the scale of the program has been expanded (from an initial $50 million budget to $1.5 billion over 10 years) and the timeline accelerated.

NSF appears to be violating a congressional directive that caps reductions to any NSF directorate at 5%. Despite an essentially flat budget (overall the proposed NSF budget is down 3% year over year), Science reports that the agency is proposing reductions of 20% – 30% or more in many programs.

Deals Deals Deals

It was a busy month for mergers and acquisitions in scholarly publishing. Notably, three of the acquisitions discussed below are by not-for-profit organizations. While a common tool in the strategic toolbox of commercial enterprises, M&A activity appears to be increasing among not-for-profits in the sector (though we are not aware of a comprehensive study on the subject).

Clarivate divests its Life Sciences & Healthcare business. The biggest story was the announcement by Clarivate that it has entered into a definitive agreement to sell its Life Sciences & Healthcare (LS&H) business to investment firm Altaris for $600 million ($500 million in cash, $25 million tied to transition services, and a $75 million seller note). According to Clarivate’s 2025 Annual Report, the LS&H division earned $390 million in revenue and $120 million in adjusted EBITDA in 2025 (down from $419 million in 2024). This prices the deal at 5 times EBITDA and less than 2 times revenues.

Clarivate’s LS&H division included assets purchased from Decision Resources Group in 2020. Six years ago, Clarivate paid $950 million for DRG, a price that reflected 12 times DRG’s adjusted EBITDA.

The stated purpose of the divestiture is primarily to pay off debt (Clarivate carries an extremely heavy debt load—over $4.3 billion USD). However, Clarivate notes the divestiture will additionally “enhance the quality of our revenue mix, lower capital intensity, and improve margins.” Which is to say the remaining business units have more similarities, better margins, and more recurring revenues.

This strikes us as a sensible move for Clarivate. The company’s debt burden considerably exceeds their market cap, and their stock is down over 77% since going public despite their business units throwing off healthy amounts of cash. Still, selling a division for $600 million that you recently purchased a portion of for $950 million must have been a difficult pill to swallow.

Annual Reviews purchases Underline Science. Annual Reviews continues its expansion into the conference business with the purchase of Underline Science for an undisclosed sum. Underline provides services and software to support scientific conferences. This includes the video capture of conference sessions, which are aggregated in the Underline Digital Video Library.

The purchase of Underline follows previous recent acquisitions by Annual Reviews, including Knowledge Unlatched, The Charleston Advisor, and The Charleston Conference.

In an interview with Research Information, Richard Gallagher, President and Editor-in-Chief of Annual Reviews, explains his thinking about how Underline Science fits in the growing Annual Reviews portfolio:

As AI reshapes knowledge discovery, the judgment of trusted experts becomes more important, not less. The reviews will combine human insight with AI capabilities, while conferences provide the communities in which knowledge is debated, refined, and extended. Together, they support the synthesis of knowledge and the progress of research.

Without knowing the purchase price, it is impossible to fully evaluate this acquisition. That said, the Underline Digital Video Library has traction among library customers. It may benefit from economies of scale (and package economics) and be sold alongside Annual Reviews journals. Underline will also, presumably, provide supporting infrastructure for The Charleston Conference. While the recent COVID-19 pandemic and geopolitical turmoil underscore the risks inherent in the conference business, Underline seems to skirt much of this risk by focusing on enduring video capture, conference platforms, livestreaming, and virtual and hybrid event services. Underline is also based in Europe, which provides some additional market diversification for US-based Annual Reviews.

Amsterdam University Press acquires ATF Press Publishing Group. Amsterdam University Press has acquired Australian ATF Publishing Group for an undisclosed sum. The deal includes ATF’s backlist of over 400 titles, with a focus on theology, religious studies and social ethics.

MDPI acquires three journals from Editores Medicorum Helveticorum (EMH). MDPI has acquired three medical titles from EMH for an undisclosed purchase price. The journals are: Swiss Archives of Neurology, Psychiatry and Psychotherapy (SANPP), Primary and Hospital Care (PHC), and Cardiovascular Medicine. 

SPIE acquires The Lens. SPIE, the applied optics and photonics society, has acquired The Lens for an undisclosed sum. The Lens is an analytics platform that combines patent analysis with scholarly research metadata. Given SPIE’s focus on applied science (sensors, lithography, lasers), this acquisition seems a good strategic fit.

Supersizing AI Legal Risk

There is a lot of talk about how AI may replace some legal work, but so far it seems to mostly be adding to the workload of trial lawyers.

The headline news from the last month was the final approval of a landmark $1.5 billion settlement between Anthropic and authors. The court has called the settlement the largest copyright class action settlement in US history. Authors whose books were pirated by Anthropic can claim approximately $3,000 per book. And as Mashable points out:

The settlement only releases Anthropic from liability for how it acquired the training data in the past—not from future lawsuits over what the chatbot generates, or new claims going forward. Judge Araceli Martínez-Olguín was explicit that the settlement does not release claims for future harm or claims “based on the output of AI models.”

In other “AI and lawsuits” news:

• A group of book publishers (Hachette Book Group, Cengage Learning, and Elsevier, along with author Scott Turow) are suing Google. The publishers allege that Google trained Gemini on books supplied for other purposes, such as participation in Google Books, Google Play Books and Google Scholar.
• 35 newspaper publishers filed suit against OpenAI and Microsoft for allegedly scraping their websites to train ChatGPT.
• A similar lawsuit was filed by News Corp against search engine Brave Software.

Perhaps most notably, a new front against AI companies has been opened: shareholder suits. Bloomberg reports on three shareholder suits that have been filed against Adobe and Microsoft that claim the companies misled investors by “concealing illicit AI practices like scraping copyrighted material to train their models.” The suits argue that such actions were a breach of executives’ fiduciary duties and in some cases even securities fraud. The suits seek compensation for harm caused to the companies (and their share prices). If successful, these suits have the potential to “supersize” legal risk already faced by AI companies—get sued by copyright holders and then get sued on top of that by shareholders.

Briefly Noted

Over 150 large tech companies have signed an open letter advocating for a light touch to regulating open weight AI models. Every major company involved in AI development, with the lone exception of Anthropic, has signed the letter. The letter was released in the wake of the launch of Moonshot’s Kimi K3, a Chinese model that approaches the capabilities of the US frontier models. Kimi K3 is an open weight model, allowing companies to download it free of charge and run it on private clouds (or theoretically on their own hardware). Companies using open models still face substantial costs of running the models, but these costs are far less than the costs associated with frontier models.

As The New York Times reports, US tech companies are working to influence government policy in favor of open models because, with the exception of the frontier model builders themselves, everyone else stands to benefit from a landscape where there are multiple lower-cost models (Nvidia can sell more GPUs if there are more model builders, Microsoft can sell software that coordinates models to control costs, Amazon can sell more compute). The shape of the AI landscape has relevance beyond tech companies, however. Organizations with deep reservoirs of high-value content are better positioned in a market with many models (and with specialized applications that employ different models for different use cases) as compared with a market dominated by two or three AI companies that supplant specialized software applications. A more diverse market both provides more licensing opportunities and more opportunity to develop specialized software products (including those incorporating lower-cost open weight models). Perhaps most importantly for readers of The Brief, it is also a landscape that is better aligned with existing institutional content licenses. Institutions, such as universities, working with various open models are more likely to need to retain licenses to publisher content, which might be accessed by such models (whereas in a world dominated by two or three models, the models themselves might directly provide access to publisher content).

The US Commerce Department has lifted restrictions on Anthropic’s most powerful models (we had noted last month in The Brief that use of the “Fable” model had been halted).

Many conversations (and some new initiatives) in the industry are focused on the question of how to better convey trust signals in a changing digital landscape at a moment when trust in organizations appears to be at a low point. The latest entry in this category is the TrustMarc Initiative (the unfortunate spelling of the term “mark” is a headwind that this project will have to push against). TrustMarc is envisioned as a kind of platform for trust frameworks and signals. From their website FAQ:

A TrustMarc makes an organization’s existing trust practices visible. It does not create new trust standard or certify quality—it surfaces the policies, governance, preservation, provenance, and other evidence of trustworthiness that already exist, so that both readers and automated systems can see them.

The TrustMarc infrastructure is managed by Coherent Digital “under the guidance of an independent Standards Council.”

Wiley has announced that it will join the US Department of Energy’s Genesis Mission Consortium. The aim of the Genesis Mission is to link together scientific assets (DOE’s National Laboratory supercomputers, various scientific datasets, automated “robotic laboratories”) and make them available to AI models in order to increase the velocity of discovery. Wiley is the first scientific publisher to join.

OpenEvidence has launched a new professional education platform. The platform will offer continuing education (CE) credits to NPI-verified physicians, nurse practitioners, and physician associates, as well as maintenance of certification (MOC) credits to eligible physicians through participating certifying boards.

How novel is that research paper? A competition to quantify the concept of “novelty” crowns a winner: LENS, which stands for “LLM-Evaluated Novelty and Significance,” is a tool that “uses cues including a paper’s cited references to summarize the state of knowledge on which the paper builds. It then evaluates whether a paper solved important problems, introduced valuable methods or concepts, or provided surprising findings that advanced the field.” As AI systems look for new signals to help with rank and relevance, novelty might be a useful signal.

A new “humanizer” tool available in GitHub is being used by some scientists to reduce signs of AI-written text. The tool has drawn mixed reviews according to Nature, with some scientists praising it for sharpening language and others concerned that it will lead to more undisclosed use of AI in writing papers and reviews. The tool’s creator, Jie Ding, a researcher at the University of Minnesota, responded to concerns: “I’d separate the tool from the behavior. The ethical issue is the non-disclosure and intent behind it, not the existence of an editing aid.”

***
AI can commoditise information, but not community. —Richard Gallagher, President and Editor-in-Chief, Annual Reviews