Breaking AI & Tech News Analyzed

The Rise of Recursive Self-Improvement in AIThe word "recursion" is the latest buzzword in AI circles. Two separate startups have taken on the name, and many more have started referencing recursive self-improvement (RSI) in their roadmaps. Like AGI before it, RSI has become a three-letter byword for a cataclysmic AI takeoff – even if there's still a little disagreement about what it exactly means.In basic terms, RSI refers to an AI system that can continuously upgrade itself. Once AI systems can manage the upgrade cycle better than humans, the process can become a closed loop, limited only by the compute power they can access, and humans are no longer necessary or even helpful.Scary or not, that's a vision that a lot of AI labs are eager to chase.Key Players Pursuing Recursive SystemsEarlier this month, well-known AI researcher Richard Socher launched the aptly named Recursive Superintelligence with RSI as an explicit goal. "Our main focus is to build truly recursive, self-improving superintelligence at scale," Socher told TechCrunch at launch, "which means that the entire process of ideation, implementation, and validation of research ideas would be automatic."A number of other prominent researchers are already chasing that same goal, hoping for a breakthrough that will make recursive self-improvement possible.One of the most prominent is Andrej Karpathy, a legendary figure from Tesla and OpenAI, who is using agent swarms to train LLMs on simple tasks for a project he calls Auto-Research. Karpathy has been unusually open about the project, tweeting about milestones regularly and making the building blocks available through a public GitHub repo. So far, the work has mostly been confined to making minor improvements on a GPT-2 scale model — as Karpathy noted in March, "It's not novel, ground-breaking 'research' (yet)" — but it's been enough to convince lots of other researchers to follow the RSI dream. And with Karpathy now working on pre-training at Anthropic, he will have plenty of opportunity to apply the idea at a larger scale.Adaption — founded by Cohere and Google alum Sara Hooker — recently launched a similar tool called AutoScientist in an effort to automate frontier training. Like Karpathy's auto-researchers, the system trains agents to make incremental improvements — but for Adaption, the goal is to make it easier to train a full-scale frontier model. If those same researchers start to push the frontier forward, the system could quickly spiral into something very much like RSI.Disarray founder Doris Xin drew more specific RSI interest when her self-trained machine learning agent took home 28 medals in a recent Kaggle competition, beating out many human-trained agents. As she sees it, the major challenge is reliability."I would argue, given infinite compute and infinite time horizon, we are already there," Xin told me. "I want to make an argument that this is not a creative endeavor, really. It's just a lot of meat-and-potatoes engineering."The Current State of Self-Improving AIThere's also plenty of evidence that the AI industry isn't very close to recursive systems in any meaningful way — and is still grappling with talking to a wary public about its progress. So Google CEO Sundar Pichai basically admitted in a recent podcast interview."It's a continuum, and we are all definitely making progress," Pichai said. "But in the way people describe RSI, that would represent a next level of acceleration and would have a lot of implications, but we aren't quite there yet."But the continuum includes an awful lot of self-improving AI systems.In January, one of Anthropic's lead programmers for Claude Code estimated that "close to 100%" of his team's code was written by the tool — a frank admission that Claude Code was literally writing itself.Just because engineers are using an AI tool doesn't mean the tool can replace them — but Anthropic seems to be getting close to replacing engineers too. In a recent survey tied to the Mythos preview, five out of 18 Anthropic engineers believed that, with harness improvements, this version of Mythos could soon substitute for an L4 engineer — a midlevel programmer who can take on involved projects without supervision.Still, there were some of the same weaknesses you might expect."Some of Claude's major reported weaknesses compared to an L4 include: self-managing week-long ambiguous tasks, understanding org priorities, taste, verification, instruction-following, and epistemics," the report reads.In other words, its weaknesses are everything involved with self-direction, which is the cornerstone for RSI. But sure, for everything else, Claude is ready to step right in.Expert Perspectives on RSI TimelinesJust like the AGI term before it, the AI industry also can't tell us how far away it is from showcasing a meaningful recursive system. When Georgetown's Center for Security and Emerging Technology assembled a group of experts to study RSI last year, the group found a major split in assessments — some expecting an imminent "superintelligence" style explosion while others expected slower progress and an eventual plateau. But all agreed that recursion made the future especially difficult to predict.Helen Toner, director of CSET and a former board member at OpenAI, told TechCrunch that simply using AI tools to do AI research isn't enough to qualify as RSI. "They're just using AI for as much as they can," Toner told TechCrunch. "And I think that is different from the classic definition of RSI, which is really that there are no humans needed."Toner pointed to a recent post by METR's Ajeya Cotra, which distinguishes different milestones on the path to the AI research takeover. One step, which Cotra calls "adequacy," would come when the system can still perform research after all humans are removed — even if the resulting research isn't as valuable or efficient. "Parity" comes when an AI-only system is as good at research as a human-only system. "Supremacy," the final stage, comes when an AI-only system outperforms a collaborative system between humans and AI.Ultimately, Cotra concludes that AI is very close to the adequacy threshold of being able to produce some work on its own — similar to the incremental changes made by Karpathy's Auto-Research system. "I wouldn't be totally shocked if you told me this milestone had already passed, and I expect it to happen in the next couple years," Cotra wrote.She was less clear on when parity will come, but once it does, she thinks it would "massively accelerate the pace of AI progress, leading to AI research supremacy within another year."The Challenges Ahead for Recursive AIWith so much of AI built on scaling laws, there's a strong tendency to think RSI will follow the same curve. Toner thinks that many of those pursuing AI research and development via RSI "think of it as a pretty smooth ladder, where you can just keep scaling up."But even if AI researchers are able to make incremental improvements like Karpathy's auto-researchers, there will be larger challenges in handing off the whole process of research. Toner put it in terms of the history of computing, which has seen human beings handing off more and more of the process while still directing things from the top."We went from machine languages to assembly language and compiled languages; you're getting further and further from the guts of the computer," Toner said. "But the human is still, in some intuitive sense, running the show."Moving beyond that paradigm will take significant challenges, both in engineering and alignment. But even with the massive investments happening, there's no infinite compute available — and the basic trade-off between human labor and machine intelligence will be hard to overcome.The Future of Recursive Self-ImprovementAs for a total recursive AI system of apocalyptic visions? The only thing researchers essentially agree on is that, like AGI, it's not here yet.

Executive Order on AI Model Review Put on HoldPresident Donald Trump announced a delay in signing the anticipated executive order that would task the Office of the National Cyber Director and other agencies with evaluating AI models for security before they are released.Details of the Delayed Order and Its Controversial ProvisionsThe order would require AI companies to share advanced models with the government 14 to 90 days prior to launch.It was motivated by concerns over recent releases such as Anthropic’s Mythos and OpenAI’s GPT-5.5 Cyber, which can quickly discover and exploit security flaws.Trump said he “didn’t like certain aspects of it” and feared the language could become a “blocker” to U.S. leadership in AI.Reports suggest the delay also stems from insufficient availability of tech CEOs to meet with officials on short notice.Potential Economic and Competitive ImplicationsMandating early model disclosure could affect the speed of innovation for U.S. firms.Companies may view the requirement as a competitive disadvantage relative to foreign rivals not subject to similar constraints.Broader Impact on U.S. AI Governance and International CompetitionThe postponement signals a tension between national security objectives and the desire to maintain a technological edge over China and other global players. It also raises questions about how future AI oversight will balance safety with market agility.What May Come Next for AI Regulation Under the Trump AdministrationAnalysts expect further revisions to the order’s language before a final signing, potentially narrowing the scope of mandatory disclosures or extending the review timeline. Ongoing dialogue with industry leaders will likely shape the final framework, influencing the trajectory of U.S. AI policy in the coming months.

The Evening’s Dual Focus: Franceschi’s Choreography and Krasner’s StoryThe Guardian’s review opens with a clear picture of the night’s two pillars: choreographer Antonia Franceschi, a former NYCB dancer turned artistic director of the New York Theatre Ballet, and the subject of the evening’s most ambitious work, abstract expressionist painter Lee Krasner, whose legacy is often eclipsed by her husband Jackson Pollock.Prophecy in Motion: A Sketch of Lee Krasner’s LifeFranceschi’s piece Prophecy (still a work‑in‑progress) is a dance‑theatre vignette that layers voice‑over excerpts from Krasner’s and Pollock’s own words over sparse, gestural movement. The choreography leans on minimalism, allowing a single hand gesture or a tender head tilt to amplify the emotional weight of the text. The reviewer notes that the work aims to evolve into a full‑length production that charts Krasner’s entire career.Numbers on the Stage: Run Dates and Audience ReachVenue: The Mount Without, BristolRun until: 22 May 2026While ticket‑sale figures are not disclosed, the limited‑run schedule suggests a targeted, high‑impact engagement with regional audiences, positioning the piece as a test‑bed before a broader rollout.Reframing the Narrative of Female Artists in DanceThe review argues that Uncaged+ and Prophecy together challenge the traditional male‑centric mythos of abstract expressionism by foregrounding Krasner’s agency, creative labor, and the domestic constraints she navigated. By juxtaposing her story with Pollock’s larger‑scale myth, the production invites a re‑examination of how dance can reinterpret art‑historical narratives and give voice to previously marginalised figures.Future Prospects: From Sketch to Full‑Length ProductionCritic Roseanna Anderson expresses a strong desire to see the sketch mature into a full‑length work, noting that the current minimalism serves as a compelling proof‑of‑concept. If the piece secures further funding and audience interest, it could become a landmark dance‑theatre exploration of a female artist’s life, potentially touring beyond the UK and influencing future interdisciplinary collaborations.

Anthropic’s Claude Mythos to be Presented to the Financial Stability BoardAnthropic will brief the Financial Stability Board (FSB), chaired by Bank of England governor Andrew Bailey, on the cyber‑defence implications of its Claude Mythos model, which has raised alarm among security experts.Mythos is not being released publicly; access is limited to select tech firms and banks such as Apple and JP Morgan.The briefing follows a report by the Financial Times and confirmation from a source familiar with the discussions.The FSB’s membership includes senior officials from the US, UK, Australia and China.Quantifying Mythos’ New Cyber‑Testing PerformanceThe UK’s AI Security Institute (AISI) noted a “notable capability jump” in the version shown to banks. In the “cooling tower” test, Mythos succeeded in 3 out of 10 attempts – a first for any model evaluated by AISI.Previous iterations had not completed the test.AISI reports that the length of autonomous cyber tasks has doubled within months.Implications for Global Financial CybersecurityThe briefing comes as the International Monetary Fund (IMF) warned that AI‑driven cyber risks are rising for financial stability. Central bank leaders, including Goldman Sachs CEO David Solomon and JP Morgan CEO Jamie Dimon, have already expressed heightened awareness of Mythos’ capabilities.Cyber risk does not respect borders; inconsistent oversight could weaken the interconnected financial system.Experts caution that most breaches still stem from traditional weaknesses such as weak authentication.What the Next Phase of AI‑Driven Cyber Risk May Look LikeAISI is developing tougher hacking tests to track AI progress, while the FSB is expected to issue recommendations for coordinated oversight among regulators. If the trend of rapid capability gains continues, financial institutions may need to embed AI‑specific cyber‑defence measures into their risk frameworks.Potential for tighter collaboration between AI developers and regulators.Increased scrutiny of AI models before deployment in critical infrastructure.

The Tech Diplomacy Mission to BeijingPresident Donald Trump is embarking on a high-stakes visit to China this week, accompanied by an impressive delegation of American tech industry leaders. The guest list reads like a who's who of Silicon Valley and corporate America, suggesting that technology will be a central focus of discussions with Chinese President Xi Jinping, though potentially following any developments regarding the situation in Iran.A-List of Tech Titans Joining the Presidential DelegationThe presidential delegation includes some of the most influential figures in American technology. Outgoing Apple CEO Tim Cook, SpaceX and Tesla CEO Elon Musk, Meta's recently appointed president Dina Powell McCormick, Micron CEO Sanjay Mehrotra, Cisco CEO Chuck Robbins, and Qualcomm CEO Cristiano Amon are all confirmed to join the president.The Notable Absence of Jensen HuangSurprisingly absent from the delegation is Jensen Huang, CEO of Nvidia - the world's most important chip manufacturer. Huang, who has close ties to Trump, previously criticized US limitations on chip sales to China in an April interview, expressing concern that a "loser mentality" could cost America its edge in AI. His absence suggests that a major semiconductor deal may be less likely, though an announcement from Micron remains possible.Cook's Diplomatic Role and Apple's China SuccessTrump's inclusion of Tim Cook likely reflects a desire for a familiar face in high-stakes negotiations. Apple's iPhone 17 has proven enormously successful in China, driving the company's quarterly earnings to record highs. Despite moving some manufacturing to India and Vietnam, Apple still produces most of its products in China. In announcing his retirement, Apple highlighted Cook's diplomatic skills, noting that his future responsibilities would include dealing with world leaders, suggesting such diplomatic visits may become a regular feature of his post-Apple career.Following the Middle East Model for Tech DealsWhether Trump's China visit will replicate the flurry of tech deals that emerged from his May 2025 Middle East trip remains to be seen. The president is showcasing America's top business leaders - products of his hands-off approach to fostering technological innovation - while his administration simultaneously appears to be taking cues from China's more stringent approach to AI governance.US Embracing China's AI Regulatory FrameworkChina's AI laws require companies to submit their models to Beijing for review on both security and political sensitivity grounds, prohibiting content that the government finds objectionable. In a similar move, the White House is increasing its involvement with American frontier AI labs. Trump is reportedly considering an executive order that would require AI companies to submit their newest models for White House review. The administration has already announced deals with major players including Google DeepMind, Microsoft, and xAI for national security reviews of their latest releases through the Center for AI Standards and Innovation (CAISI) at the Department of Commerce.Pentagon's Standoff with AnthropicThe relationship between the Pentagon and AI startup Anthropic continues to face challenges in court, as the startup expresses concerns about military applications of its technology while the Pentagon has designated the company as a supply chain risk. Vice President JD Vance has requested that Anthropic not expand access to its powerful cybersecurity-focused model Mythos beyond its initial list of partners, according to the Wall Street Journal, highlighting the growing tensions between AI innovation and national security concerns.

In just three months, AI‑powered hacking has moved from a nascent problem to an industrial‑scale threat, according to a Google threat‑intelligence report released on May 11, 2026.Scale and Sophistication of AI‑Assisted ExploitsThe report documents that criminal syndicates and state‑linked actors from China, North Korea and Russia are leveraging commercial models—including Gemini, Claude and tools from OpenAI—to automate vulnerability discovery, craft malware and conduct rapid, large‑volume attacks. Notable findings include:A criminal group on the brink of a “mass exploitation” campaign using an unnamed LLM.Experiments with OpenClaw, an AI agent that can automate extensive user data handling and even mass‑delete email inboxes.Anthropic’s decision to withhold its newest model, Mythos, after it identified zero‑day flaws across every major OS and web browser.Financial and Operational Stakes Highlighted by Recent FindingsWhile the UK government projects a £45 billion boost in public‑sector savings and productivity from AI, the Ada Lovelace Institute (ALI) warns that many of these figures rest on untested assumptions. The ALI report highlights gaps such as:Reliance on time‑saving metrics rather than service‑quality outcomes.Insufficient accounting for employment impacts in the public sector.Short‑term study windows that miss long‑term productivity trends.Implications for Cybersecurity Policy and Industry DefencesGoogle’s findings underscore the need for coordinated defensive action across the industry. Recommendations include:Mandating early‑stage impact measurement for AI deployments in government departments.Supporting longitudinal studies that track AI‑driven productivity over years, not weeks.Encouraging transparency around the use of LLMs in both offensive and defensive security tools.Outlook: How the Threat Landscape May EvolveExperts like Steven Murdoch of University College London note that the traditional bug‑discovery process is already being supplanted by LLM‑assisted methods, suggesting a prolonged period of adjustment for defenders. As AI models become more capable, the balance between accelerated attack capabilities and defensive innovation will likely dictate the next wave of cyber‑risk management strategies.

The Emergence of Mythos AI Anthropic's recent announcement about its new model, Claude Mythos Preview, has raised both excitement and concern. The model is remarkably effective at finding security vulnerabilities in software, but Anthropic has decided not to release it to the general public. Instead, it will only be available to a select group of companies to scan and fix their own software. The Capabilities of Mythos AI While Anthropic's model is impressive, it's not unique. Other models, such as OpenAI's GPT-5.5, have comparable capabilities. The UK's AI Security Institute found that GPT-5.5 can also find software vulnerabilities. Additionally, smaller and cheaper models have been able to reproduce Anthropic's published results. The Financial Implications of Mythos AI The high cost of running Mythos AI is a significant factor in Anthropic's decision not to release it publicly. The company's valuation can be boosted by hinting at the model's capabilities without actually proving them. This strategy allows Anthropic to maintain a competitive edge while limiting access to the model. The Impact on Cybersecurity The emergence of models like Mythos AI has significant implications for cybersecurity. These models can be used by both attackers and defenders to find and exploit vulnerabilities in software. This could lead to a more dangerous and volatile world, with increased risks of cyber attacks and data breaches. The Future of AI and Cybersecurity As AI models continue to improve, we can expect to see more frequent software updates and a greater emphasis on cybersecurity. However, the long-term implications of these models are more complex. They may be used to find loopholes in complex systems, such as tax codes and regulatory systems, which could have far-reaching consequences for society. The Broader Implications of Mythos AI The capabilities of Mythos AI have broader implications beyond cybersecurity. These models can be used to analyze complex systems and find vulnerabilities, which could be applied to areas such as tax law and environmental regulations. This raises important questions about the potential misuse of these models and the need for careful consideration of their development and deployment.

The US Government's Access to AI Models Tech giants Microsoft, Google, and xAI have agreed to allow the United States federal government access to their new artificial intelligence models for national security testing. The Center for AI Standards and Innovation (CAISI) Agreement The Center for AI Standards and Innovation (CAISI) at the Department of Commerce announced the agreement on Tuesday amid increasing concerns about the capabilities that Anthropic’s newly unveiled Mythos model could give hackers. The Data Analysis and Testing Under the new agreement, the US government will be allowed to evaluate the models before deployment and conduct research to assess their capabilities and security risks. Microsoft will work with US government scientists to test AI systems “in ways that probe unexpected behaviors”. The Impact Analysis on National Security Concern is growing in Washington over the national security risks posed by powerful AI systems. By securing early access to frontier models, US officials are aiming to identify threats ranging from cyberattacks to military misuse before the tools are widely deployed. The Future Outlook and Implications The move builds on 2024 agreements with OpenAI and Anthropic under President Joe Biden’s administration. CAISI, which serves as the government’s main hub for AI model testing, said it had already completed more than 40 evaluations, including on cutting-edge models not yet available to the public.

The Power of Anthropic's Mythos Model When Anthropic unveiled its new Mythos model in April, it also delivered a stern warning to anyone developing software. The model was so powerful at sniffing out software vulnerabilities, the lab claimed, that it had discovered thousands of high-severity bugs that would need to be fixed before it could be made public. Improving Software Security with AI Now, security researchers for Mozilla's Firefox browser are providing a closer look at what that process has looked like in practice, and what Mythos' powers mean for software security at large. In a post published on Thursday, Mozilla said Mythos has unearthed a wealth of high-severity bugs, including some that had lain dormant in the code for more than a decade. The Data Behind the Discovery In April 2026, Firefox shipped 423 bug fixes, compared to just 31 exactly a year earlier. The researchers have also published details on 12 of the bugs, which range from a pair of unusual sandbox vulnerabilities, to a 15-year-old error in how the browser parses an HTML element. The Impact on Cybersecurity The fact that the system helped reveal vulnerabilities in Firefox's 'sandbox' system is particularly impressive, given how intricate an attack that exploits it needs to be. To find sandbox vulnerabilities, the model must write a compromised patch for the browser, then attack the most secure part of the software with the new code implemented. Finding and demonstrating the bug is a delicate, multi-step process, requiring both creativity and close attention. The Future of AI in Cybersecurity It's still not clear how AI's emerging capabilities will change the broader balance of power in cybersecurity. One month since Mythos was previewed, most of the bugs discovered likely haven't been patched, which makes it hard to capture the full scope of their impact. Anthropic has been scrupulous about following responsible disclosure norms, but it's likely bad actors are using similar techniques behind the scenes, even if the models they're using aren't quite as good. The Prediction Speaking at a recent event, Anthropic CEO Dario Amodei was optimistic that the new tools would ultimately favor defenders. 'If we handle this right, we could be in a better position than we started, because we fixed all these bugs. There are only so many bugs to find,' Amodei said. 'So I think there's a better world on the other side of this.'