Flavia Rotondi / Bloomberg: The Vatican says Anthropic co-founder Christopher Olah will join Pope Leo on May 25 to launch the pope's first encyclical, setting out his views on the AI age  —  Pope Leo XIV's launch of his first encyclical focusing on artificial intelligence next week will feature an unexpected guest: Christopher Olah, co-founder of Anthropic.

95% of enterprise AI pilots fail to launch. Why? The post Why Your AI Demo Will Die in Production appeared first on Towards Data Science.

Robbie Whelan / Wall Street Journal: Decart, which offers real-time generative video and GPU optimization tech, raised $300M at a ~$4B valuation, up from $3.1B after raising $153M in August 2025  —  Decart's valuation hits nearly $4 billion as investors pour capital into startups making AI computing more efficient

Discussion | Link

Discussion | Link

Britt Paris's new book 'Radical Infrastructure: Imagining the Internet from the Ground Up' tells the story of the physical internet, and how it can benefit people, not corporations.

“With your permission, your child’s lead teacher may wear a small teacher-worn camera that captures the teacher's approximate first-person perspective, and/or we may place a fixed video camera in the classroom,” a document given to parents and later shared with 404 Media reads.

James Hunt / The Block: Strategy acquired 24,869 bitcoin for ~$2.01B at an $80,985 average price from May 11 to May 17, taking its holdings to 843,738 BTC, or 4%+ of the total supply  —  Quick Take  — Strategy has purchased another 24,869 BTC for approximately $2.01 billion at an average price of $80,985 per bitcoin …

Julia Love / Bloomberg: Sources: competition for Google's TPUs among its employees and researchers intensified as it prioritizes cloud customers and flagship AI products over research  —  In the race to build the infrastructure that powers artificial intelligence, Alphabet Inc.'s Google has an enviable position …

AI agents capable of discovering and exploiting obscure vulnerabilities are emerging alongside developers producing vast amounts of potentially flawed AI-generated code, forcing defenders to adapt accordingly.

Table of Contents LLM Observability with Self-Hosted Langfuse and vLLM Introduction to LLM Observability with Langfuse How Langfuse Fits into an LLM Observability Stack Langfuse Architecture for LLM Observability Why Understanding LLM Observability Architecture Matters Setting Up a Self-Hosted Langfuse… The post LLM Observability with Self-Hosted Langfuse and vLLM appeared first on PyImageSearch.

Chris Metinko / Axios: Dust, which helps enterprises design and deploy specialized AI agents that work alongside humans, raised a $40M Series B led by Abstract and Sequoia  —  Dust, a human-agent collaboration platform, raised a $40 million Series B led by Abstract and Sequoia, CEO Gabriel Hubert tells Axios Pro exclusively.

The volume of social media posts makes content moderation challenging – especially when it comes to more subtle hate speech. Peter Dazeley/The Image Bank via Getty Images This article includes examples of antisemitic hate speech. The men accused of carrying out high-profile antisemitic attacks in the United States in recent years shared an important characteristic: They posted hate speech on their social media accounts beforehand. The FBI said the man who drove his truck into a synagogue outside Detroit in March 2026 posted on Facebook that “Israel is a cancerous/malignant growth” and “Israel is pure evil.” The online footprint of the gunman charged with shooting and killing two Israeli Embassy staffers at the Capital Jewish Museum in May 2025 contained anti-Israel comments. The shooter sentenced to death for killing 11 worshippers at the Tree of Life synagogue in Pittsburgh in October 2018 frequently used antisemitic hate speech in his social media. Hate speech uses feelings, emotions and attitudes that seek to dehumanize individuals or groups. At times, animosity is clear. But it can also take a more hidden form, using code words or terms understood only by like-minded people. Coded hate speech can evade online content censors and recruit people who might balk at more clearly discriminatory speech. There are an estimated 5.7 billion social media accounts worldwide. Even when hate speech is explicit, content moderators struggle with the volume and deciding how much to monitor users’ speech. There are also alternative – some argue extremist – sites that limit content moderation, including 4chan, BitChute, Gab, GETTR, Parler, Rumble and Truth Social. We are a group of interdisciplinary researchers at American University who study the rhetorical strategies behind overt and coded hate speech on social media. Our Unmasking Antisemitism project uses artificial intelligence, qualitative analysis and survey experiments to develop studies and tools to detect both types of terms. This article discusses examples of antisemitic hate speech that are disturbing but illustrate types of terms and how to counter this dangerous influence. Two types of hate speech To understand the difference between direct and coded hate speech, consider shooter Robert Bowers’ language before the Tree of Life massacre. On Gab, he used older, virulently antisemitic slurs such as “kike,” a “highly offensive term used to insult and denigrate people of Jewish faith or ethnicity that is widely considered to be a form of hate speech,” according to the American Jewish Committee. A fence outside the Tree of Life synagogue, site of the 2018 mass killing, holds artwork from schoolchildren on April 21, 2003, in Pittsburgh, Pa. Jahi Chikwendiu/The Washington Post via Getty Images Other extremist terms are just as offensive but less obvious, such as “oven dodger,” which Bowers also used on Gab: a reference to how German Nazis systematically exterminated Jews during the Holocaust. Like overt phrases, coded terms often draw on older, well-researched antisemitic tropes, such as “Jews have too much power,” repacking them in new words and phrases. They can also have double meanings, which makes hate speech harder to moderate. The original definition of “globalist” refers to a person who believes that policies should be planned with the whole world’s interest in mind rather than just one country. But globalist also has an antisemitic connotation. As the American Jewish Committee “Translate Hate” glossary puts it, antisemites often use “globalist” to disparage Jews, promoting a conspiracy theory that “Jewish people do not have allegiance to their countries of origin, like the United States, but to some worldwide order – like a global economy or international political system – that will enhance their control over the world’s banks, governments and media.” This repackages long-standing Nazi and Soviet propaganda about Jews based on historical antisemitic tropes. How terms develop In the early days of social media, companies responded to criticism of the more hateful content on their platforms by using a combination of AI and human analysis to moderate content. The automated tools use natural language processing to analyze context, detect slurs and flag content. Human workers analyze more complex language, such as irony and slang. Content moderators work at a Facebook office in Austin, Texas, in 2019. Ilana Panich-Linsman for The Washington Post via Getty Images But keeping up with the volume of posts is challenging, especially for more subtle hate speech. Our team’s goals are to identify coded antisemitic terms, understand how they develop, and create technology to track them. The key is to understand that hate terms have a life cycle. Some take a path toward more public use, while others disappear. New terms tend to emerge from a small set of people considered leaders or influencers in antisemitic circles online. In some cases, their communities adopt the term and normalize it; other times, it’s dropped from use. The term “cultural Marxism,” which has its origins in the antisemitic belief that Jewish intellectuals seek to subvert Western culture, was adopted into wider use. “Jew jab,” on the other hand – a white supremacist conspiracy theory claiming that COVID-19 vaccines were a Jewish plot to harm people – soon disappeared. Tracking hate In our initial pilot project we started with 46 antisemitic terms, both overt and coded, from the American Jewish Committee’s glossary. We entered the terms in Pyrra, now called Alert Media – a private software company that allows users to scrape posts from a collection of social media sites. Researchers trained in definitions of antisemitism, historical antisemitic tropes and hate speech detection identified 24 additional terms. White supremacists use the symbol “1488,” for example, to identify each other. The first part, “14,” references the “14-word” slogan of white supremacist leader David Lane. The “88” stands for “Heil Hitler,” based on “h” being the eighth letter of the alphabet. Other coded terms are less well known, such as “DOTR” or “Day of the Rope,” a reference to the 1978 book “The Turner Diaries,” which was written under a pseudonym by neo-Nazi William Pierce. To track which coded terms have spread to the general public, we scrutinized mainstream media content and ran survey experiments to see whether people recognized them. We also developed an AI software tool designed to automatically track how coded language evolves. The app is trained on data from Pyrra and learns to identify new antisemitic terms based on the context in which they appear. First, the app identifies distinctive terms based on how frequently they appear in each post, versus how rare they are on the platform in general. To find out whether these terms have an antisemitic connotation, we encode their context, such as other words in the post, and calculate whether it is close to the context of already known antisemitic terminology. Some of the terms our app has identified are explicit, while others are coded. This approach can also be applied to hate speech targeted at other groups, such as Latinos, LGBTQ+ people and women. We aim to create a tool kit that can be distributed to nonprofit groups, think tanks and policymakers considering legislative efforts to curb hate speech. Humans and machines Given the massive number of posts on social media every day, our work illustrates how detecting new hate speech requires an interdisciplinary group of researchers working with machines. One academic discipline working independently is too siloed, and humans alone can’t handle the scale. But machines alone can’t understand sophisticated human language, slang or context. History shows that at every moment of profound technological change in our communication systems, incidents targeting Jews or other minority groups go up dramatically. This era’s technical innovation is unprecedented – but unfortunately, hate speech now travels around the globe almost instantly. Technology may be part of the problem, but its immense power can be harnessed to create a solution. We received internal funding for this project from American University as part of its Signature Research Initiative.

The struggling footwear company Allbirds, which announced in April 2026 that it was rebranding into an AI company, may be one of the most recent notable examples of 'AI washing.' Business Wire Across corporate earnings calls, investor presentations and marketing pitches, “artificial intelligence” has become the buzzword of choice. Yet a troubling pattern lies under the hype. Many claims vastly overstate actual AI sophistication, misleading people about true capabilities, future outcomes and potential harms. A case in point is the recent 600% share price surge of Allbirds, after the once-trendy sustainable footwear business issued a vague announcement in April 2026 that it would pivot to AI. In the coming months, the company plans to rename itself NewBird AI and give up its status as a public benefit corporation. As a scholar who studies corporate sustainability, I see parallels between this “AI washing” phenomenon – when companies oversell the benefits of AI while glossing over the risks – and the greenwashing trend in the recent past, when companies claimed to commit to sustainability but didn’t enact fundamental change. Widespread deception was rampant, with businesses spending far more on green marketing than on actual sustainability improvements. And those efforts often backfired on both the companies and the communities they served. Even more worrisome: AI washing’s rapid rise and widespread adoption will likely eclipse the greenwashing trends. How we got here AI washing is thriving because companies and policymakers ignore four important principles. These shortfalls, in the past, also characterized greenwashing. First, AI guidelines lack standardization. By 2019, 84 sets of AI ethics principles and guidelines had already been published. By 2023, this number had exploded to more than 200 – a mess of voluntary frameworks from companies, research institutions and public organizations. Making matters worse is that the U.S. currently relies on fragmented AI rules, with most being voluntary. The Trump administration has generally sided with Big Tech to push back efforts at state or federal regulation. At a global level, one of the few exceptions is the European Union AI Act, perhaps one of the most comprehensive frameworks, but its implementation won’t be fully phased in until 2027 or later. In the early 2000s, corporate sustainability faced a similar credibility crisis. Every company measured sustainability differently, making comparisons impossible and greenwashing easy. The breakthrough came only when standardized, industry-specific metrics allowed meaningful benchmarking. Initiatives such as the Global Reporting Initiative and Sustainability Accounting Standards Board established common metrics for measuring environmental impact, social responsibility and quality of governance, known by the shorthand ESG. The U.N. climate change summits, like this one in Brazil in 2025, have offered a global forum for policymakers and business leaders on climate and sustainability issues. AP Photo/Fernando Llano When companies must report carbon emissions using the same methodology, for example, or disclose labor conditions using identical categories, investors can compare performance, identify laggards and allocate capital accordingly. This push made comparisons possible and deception harder, although it still wasn’t foolproof. For example, a 2023 United Nations Environment Programme report on the fast-fashion industry found that many companies continue to make “vague and inflated” sustainability claims. Second, there are no comprehensive frameworks in the U.S. that require businesses to judge how AI affects them in a material way and publicly disclose those impacts. Examples of AI-driven material impacts include whether algorithmic bias shapes business outcomes, or whether decisions on how to use AI systems carry significance for shareholders and the public. Instead, AI governance remains dominated by the narrow inner circle of companies that build the AI systems, while affected communities rarely have a say in determining which AI impacts are material enough to warrant public attention. For example, Big Tech companies like Google, Microsoft, Apple, NVIDIA and others adhere to their own AI governance guidelines, with relatively little public input. The development of sustainability principles offers some examples of how to build these frameworks. The EU’s Corporate Sustainability Reporting Directive requires over 50,000 companies to formally evaluate which sustainability topics are material to their stakeholders, and then disclose that information. These efforts try to ensure that accountability is clear across entire supply chains. While nowhere nearly as comprehensive, U.S. regulations such as the 2010 Dodd-Frank financial reform and California’s law requiring reporting on statewide greenhouse gas emissions provide a similar blueprint that U.S. policymakers could build on if they chose. A third problem is the general lack of third-party verification, making AI washing trivially easy. Effective disclosure means reporting all material impacts – not just cherry-picked successes. In practice, AI audits can vary dramatically in rigor, scope and methodology. One auditor might conduct extensive testing across demographic groups, analyze decision-making and validate the quality of training data. Another might simply review documentation and accept company explanations at face value. Given the variety of AI auditing models out there, different auditors may use incompatible methodologies, making results impossible to compare. If companies adopted third-party accreditation systems to assess how they use AI, they would help ensure the accountability that self-reported claims cannot match. By contrast, there was reasonable progress in this respect as companies adopted ESG principles. For example, institutions such as the Carbon Disclosure Project and Global Reporting Initiative have a network of partners that offer independent verification. These providers, certified under international standards, verify corporate sustainability data against rigorous criteria. That way, they provide the assurance that lets companies show the progress needed to unlock sustainable finance and mitigate legal risks. Third-party audits are far from perfect, but they offer a clear path for improvement. The fourth principle is robust enforcement. Early ESG initiatives relied on reputational pressure and stakeholder goodwill – things that corporations routinely ignored when profits were at stake. When change came, it was because regulations established legal liability and financial penalties. These consequences changed how corporations assess risk and continue to shape sustainability practices today. Volkswagen’s 2015 ‘Dieselgate’ scandal, for example, cost the company over US$30 billion in fines, settlements and criminal charges after U.S regulators found that the carmaker was cheating emissions tests. BP faced billions in penalties and liabilities for the 2010 Deepwater Horizon disaster, the biggest oil spill in the history of marine oil drilling operations. The current enforcement gap in AI creates a predictable dynamic. The expected value of AI washing – like potential investment gains, competitive advantage, and market valuation increases – far exceeds the expected cost in terms of penalties and risk of detection. Until enforcement imposes consequences that exceed benefits, AI washing will persist as a rational business strategy rather than a risk to a business’s reputation. Fortunately, investors are beginning to step up. The Federal Trade Commission, for example, launched Operation AI Comply in 2024, targeting deceptive AI claims, although this effort has been partially scaled back by the current Trump administration. New standards for a new era Until businesses address these four principles, AI washing will continue. Without standards and audits, even well-intentioned companies can’t know if their work meets adequate rigor. Without assessments of material impact, some groups of consumers or shareholders will be hurt. And without liability, even thorough auditors won’t be able to identify whether a business’s claims about AI are truthful. These principles, applied broadly, also help explain why greenwashing persists. For example, the lack of universal reporting standards continues to leave some gaps, with data-quality issues persisting even as reporting frameworks emerged. More fundamentally, political buy-in for ESG has diminished significantly, particularly in the U.S., where over 150 bills were introduced across multiple states by 2023 to disincentivize firms from adopting ESG. Major financial institutions – including JP Morgan, State Street, BlackRock and PIMCO – have retreated from their earlier climate commitments amid political pressure as well as antitrust concerns. This trend shows that even well designed accountability measures require durable political support to succeed. After all, corporate sustainability took more than 25 years to develop from an initial framework to mandatory standards, and it still remains a work in progress. AI, by contrast, is advancing exponentially in terms of its reach and societal impact. There may not be 25 years to catch up – but at least there are lessons from the recent past. Suvrat Dhanorkar does not work for, consult, own shares in or receive funding from any company or organization that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Musk v. Altman week 3: Musk and Altman traded blows over each other’s credibility. Now the jury will pick a side. In the final week of the Musk v. Altman trial,…

From the MGM and Caesars fiasco and MOVEit's patch nightmare to epic business blunders and the jaded reality of living in a post-breach world, Dark Reading looks back at the mistakes, miscalculations, systemic failures, and cringeworthy moments that still have us shaking our heads.

Seventy to eighty percent of patients with transthyretin amyloid cardiomyopathy (ATTR-CM), a progressive...

Learn how to get the most out of OpenAI's coding agent The post How to Maximize OpenAI’s Codex appeared first on Towards Data Science.

Ashley Gold / Axios: Steve Bannon and 60+ Trump allies sign a Humans First-led letter urging Trump to mandate government testing and approval of powerful AI models before release  —  A group of more than 60 loyal allies of President Trump is urging him to test and approve the most powerful AI models before they're released …

While deal counts remain robust for quantum startup investments and big rounds are still getting done, overall funding is on track to decline some from last year’s peaks.