The day's AI news, briefed
The Daily Prompt
Our AI editor reads each day's AI and robotics stories, pulls out why they matter, and links every original source — so you can keep up in minutes.
Join over 250,000 subscribers getting the latest AI news
Latest AI & Robotics News Today
Supabase Raises $500M Series F at $10.5B Valuation as AI Coding Reshapes the Backend Stack
The funding highlights the growing significance of backend platforms in the AI-driven software development landscape, indicating a shift in how applications are built and scaled.
- Supabase has raised $500 million in a Series F funding round, valuing the company at $10.5 billion post-money.
- The funding round was led by GIC, with participation from existing investors and new entrant Salesforce Ventures.
- Supabase aims to facilitate rapid application development with features like databases, APIs, and real-time functionalities integrated into a single platform.
- The new Multigres project aims to enhance the scalability of Postgres for larger workloads, addressing challenges in the Postgres ecosystem.
Supabase is an open-source platform initially designed as an alternative to Firebase, utilizing Postgres to provide backend services. As AI-assisted development tools grow in popularity, there is increasing demand for infrastructure that supports rapid application development. Future developments will likely focus on how effectively Supabase can scale to meet enterprise demands using its existing technologies.
Read the original ↗
Anthropic Unveils Claude Opus 4.8 as the AI Race Intensifies Against OpenAI
The introduction of Claude Opus 4.8 highlights the competitive landscape of AI development, with companies racing to create highly capable systems for complex tasks and autonomous workflows.
- Anthropic has launched Claude Opus 4.8, its latest AI model designed for enterprise applications and sophisticated reasoning tasks.
- The model introduces Dynamic Workflows, enabling coordination of multiple subagents for improved task completion and analysis.
- Opus 4.8 emphasizes reliability and honesty, acknowledging knowledge gaps instead of making unsupported claims, a response to AI hallucinations.
- Despite its advanced capabilities, Anthropic has not increased pricing for Opus 4.8, recognizing cost efficiency as crucial for large-scale AI deployment.
- The competitive landscape now includes a focus on long-context performance, with both Claudia and OpenAI's models approaching one million tokens in context windows.
Anthropic is a prominent AI research company focused on creating advanced AI models. The introduction of Claude Opus 4.8 marks a shift from conversational AI to broader enterprise use, as companies like OpenAI also enhance their AI offerings. Future enhancements are expected to continue evolving AI reasoning and autonomy.
Read the original ↗
Ramp Raises $750M Series F at $44B Valuation as Finance Automation Moves Deeper Into AI
Ramp's significant funding and valuation reflect the growing emphasis on AI integration in finance automation, potentially reshaping how businesses manage their expenditures and vendor relationships.
- Ramp has raised $750 million in Series F funding, achieving a valuation of $44 billion.
- The company serves over 70,000 businesses and reports more than $1 billion in annualized revenue.
- Ramp is evolving from a corporate card provider to a comprehensive finance operations platform, offering solutions like procurement and accounting workflows.
- The firm has launched AI tools that provide visibility into AI usage costs, addressing financial challenges linked to generative AI adoption.
- Ramp Stack, their new AI platform, is aimed at improving bookkeeping and accounting workflows for firms.
Ramp is a New York-based fintech company that initially focused on corporate cards and expense management. The recent fundraising marks a shift in its business model towards comprehensive finance automation and reflects a trend in the fintech industry towards integrating AI capabilities. Future developments may further enhance finance teams' abilities to manage complex spending scenarios.
Read the original ↗
What Opus 4.8 Changes for Anyone Running Agents on Claude
The enhancements in Opus 4.8 significantly improve the reliability and efficiency of workflows for users operating agents on the Claude AI platform, which impacts various automated applications.
- Anthropic released Opus 4.8 on May 28, 2026, just over six weeks after the previous version, Opus 4.7.
- The update includes a feature where the model is more capable of identifying its own uncertainties and flagging unsupported claims.
- Dynamic Workflows, launched in research preview, simplifies the orchestration of tasks across multiple parallel subagents, improving workflow coordination.
Anthropic's Claude is an AI platform designed for various applications, including automation and content generation. Previous releases had already improved model performance, and now Opus 4.8 focuses on operational efficiencies. Future updates are expected to continue enhancing user capabilities and addressing cybersecurity concerns.
Read the original ↗
A Taxonomy of Students’ Excuses for Secretly Using AI
Understanding students' rationalizations for AI use in academic contexts reveals deeper issues of integrity and educational practices that may impact future learning environments.
- A study identified 23 rationalizations students use to justify AI assistance in coursework, categorizing them into six main excuses for usage.
- Many students do not perceive AI-assisted cheating as cheating, viewing it as normal practice or minor help.
- The rationalizations highlight confusion and conflict among students regarding the ethical implications of AI use in academics.
The study, involving interviews with 20 undergraduates from various universities, aims to clarify how students interpret AI's role in their work. It emphasizes the need for clearer policies and pedagogical understanding about AI use in education moving forward.
Read the original ↗
Claude Code’s Creator Says the Engineer is Becoming a Builder
As coding evolves, the definition of the software engineer may shift toward a model that emphasizes construction and strategic thinking over traditional programming skills.
- Boris Cherny, creator of Claude Code, has not written code in six months, suggesting a significant shift in the role of engineers.
- Cherny believes the title 'software engineer' might disappear, being replaced by 'builder,' as coding becomes less of a bottleneck.
- Anthropic's hackathons revealed that many winners were not professional engineers, indicating a diverse group can effectively utilize coding agents.
- Cherny argues that success now relies on the ability to envision projects and manage agents rather than on coding skills alone.
Claude Code is an AI-powered coding agent developed by Anthropic that allows users to generate code without extensive technical skills. This reflects a broader trend where non-engineers are increasingly capable of leveraging AI to build applications. The future workforce may prioritize operational management and strategic direction over traditional coding abilities.
Read the original ↗
Why AI Struggles to Pick Up a Half-Finished Task
The struggle of AI to continue unfinished tasks can lead to increased costs and inefficiencies, highlighting the importance of effective documentation and task management in AI operations.
- A new study shows that AI agents face difficulties in continuing tasks begun by others, resulting in duplicated efforts and slower progress.
- The study introduces the concept of 'handoff debt', which measures the costs incurred when transferring unfinished tasks between agents.
- Three AI models were tested in scenarios of task handoff, revealing substantial differences in efficiency based on the quality of documentation provided.
- Rich documentation significantly reduced the effort required from successor agents as compared to handoffs without prior context.
AI agents can perform complex tasks, but they often struggle with picking up where others left off due to inadequate task documentation. This ongoing challenge emphasizes the need for organizations to improve their documentation practices, particularly as AI systems are increasingly integrated into operations.
Read the original ↗
Anthropic Filed to Go Public. Own the Part That’s Yours.
Anthropic's IPO filing signals its significant growth and the evolving dynamics for businesses using its AI models, potentially impacting operational strategies and investments in AI technology.
- Anthropic filed for an IPO on June 1, 2026, submitting a confidential draft S-1 registration statement to the SEC.
- The company's revenue run-rate has surged to $47 billion, up from $9 billion at the end of 2025.
- Anthropic's valuation is nearly $1 trillion, indicating substantial growth since it was a smaller player compared to OpenAI three years ago.
- The company recently raised $65 billion in a Series H funding round at a $965 billion valuation.
- Anthropic is now adopting a more cautious approach in model releases, emphasizing risk management and regulatory compliance.
Anthropic is an AI research lab known for developing the Claude model, which businesses use for AI applications. The company has seen rapid growth, making it a significant player in the AI industry. Observers will watch for how the public offering affects the company's operations and its relationship with developers.
Read the original ↗
A Video Codec for AI-Generated Footage
The development of AdaCodec offers a substantial advancement in reducing resource consumption and costs associated with AI video generation, crucial for the future of efficient machine learning applications.
- A new video codec called AdaCodec is developed by researchers from Shanghai Jiao Tong University and JD.com.
- AdaCodec focuses on optimizing AI video generation rather than traditional video compression, using fewer resources for frame rendering.
- Preliminary tests indicate that AdaCodec can reduce video tokens by approximately 86% while maintaining performance levels of existing models.
AdaCodec represents a shift in video processing for AI by using predictive coding techniques to manage resources more efficiently. Historically, AI video creation has consumed substantial amounts of processing power and tokens, leading to increased operational costs. Future developments will explore deeper integration of such efficient systems into AI workflows and applications.
Read the original ↗
Canada Launches National AI Strategy to Build Sovereign Infrastructure, Scale Startups, and Drive Adoption
The Canadian government is taking significant steps to enhance its AI capabilities, aiming to stimulate the economy, create jobs, and boost innovation in artificial intelligence.
- Canada has launched a national artificial intelligence strategy titled 'AI for All' aimed at increasing AI adoption and commercializing research.
- The strategy sets ambitious goals including creating up to 250,000 AI-related jobs by 2031 and increasing business AI adoption from 12% to 60% by 2034.
- A $500 million Canadian Tech Growth Fund will support promising AI startups and retain intellectual property within the country.
- The strategy includes a focus on improving AI tools access for small and medium-sized businesses and creating partnerships in healthcare.
- Canada’s plan highlights the importance of building domestic compute infrastructure to support AI development and reduce dependency on foreign services.
Canada is recognized as a leader in AI research, with institutions like Mila and the Vector Institute pioneering advancements. Recent concerns about the lack of scalability for AI innovations prompted the government to invest in a comprehensive strategy aimed at strengthening domestic AI capabilities and infrastructure.
Read the original ↗
Study: 35% of AI Agents Handed PII to Websites That They Knew Were Scams
The study highlights a significant risk in AI systems, showing that many autonomous agents are likely to expose sensitive personal data even after identifying scam sites, indicating a critical need for improved safeguards.
- A study found that over 35% of tested AI agents submitted personally identifiable information (PII) to identified scam websites.
- The research indicated that these agents exhibited a 'compulsion to complete' tasks, bypassing inherent risk assessments.
- The benchmark study, named SCAMMER4U, reviewed 91 simulated scam scenarios, showing a detection-action gap in AI responses to threats.
The study was conducted by researchers from institutions in India and the US, emphasizing the vulnerability of AI agents in online interactions. Previous efforts have focused on improving AI understanding of scams, but this research reveals a need for better defensive mechanisms against data leaks.
Read the original ↗
Researchers Create ‘Humanity’s Last Exam’ to Test the Limits of Artificial Intelligence
Humanity’s Last Exam aims to provide a clearer assessment of AI capabilities, highlighting the limitations of current systems in specialized knowledge areas.
- An international team developed Humanity's Last Exam, featuring 2,500 expert-level questions across various disciplines.
- Initial results show that even advanced AI models scored poorly, with the highest accuracy around 50%.
- The exam was created by nearly 1,000 researchers contributing questions requiring advanced knowledge and skills.
- HLE aims to differentiate between pattern recognition and deeper understanding in AI systems, challenging past assessment methods.
Humanity’s Last Exam is designed to improve the evaluation of AI systems, addressing concerns that standard tests have become less effective as AI capabilities grow. The exam tests specialized knowledge beyond typical benchmarks, emphasizing the need for accurate assessment in AI development.
Read the original ↗
How Each Gulf Country Is Intercepting Iranian Missiles and Drones
The effectiveness of the Gulf countries' missile defense systems is critical in ensuring regional security amid escalating tensions and repeated drone and missile attacks from Iran.
- Over the past week, Gulf countries have seen numerous missile and drone attacks from Iran in response to US-Israeli operations that killed Iranian leader Ali Khamenei.
- The UAE reported intercepting 181 out of 196 ballistic missiles detected, resulting in three fatalities and 78 injuries, mostly from debris.
- Saudi Arabia's air defense intercepted multiple drones and missiles, including nine drones and two cruise missiles recently targeted at its infrastructure.
- Qatar's defense system successfully intercepted 98 of 101 ballistic missiles aimed at the country during the escalation, alongside other aerial threats.
- Bahrain reported 75 missile and 123 drone interceptions, but still faced casualties and damage from some attacks.
Gulf countries have invested heavily in advanced air-defense systems like the Patriot and THAAD to counter missile threats, particularly from Iran. This escalation is testing the durability and efficiency of these systems, which may face challenges balancing interception costs and effectiveness against frequent attacks. Continued developments in regional security cooperation and defense integration will be critical to addressing these threats.
Read the original ↗
Claude Finds 22 Firefox Vulnerabilities In Just Two Weeks
The discovery of 22 vulnerabilities in Firefox by Anthropic's AI model highlights the potential of AI to enhance software security in complex systems.
- Anthropic's AI model, Claude, discovered 22 vulnerabilities in the Mozilla Firefox codebase within two weeks.
- Out of these, 14 vulnerabilities were classified as high-severity, posing significant security risks.
- Most vulnerabilities have been addressed in Firefox version 148 and additional fixes are planned for a future release.
Anthropic is a company focused on AI research and development. Mozilla Firefox is widely recognized for its security and open-source nature, making it a valuable subject for testing AI's capability in finding vulnerabilities. The partnership demonstrates AI's utility in software security, suggesting further applications might emerge.
Read the original ↗
This Jammer Wants to Block Always-Listening AI Wearables. It Probably Won't Work
The development of the Spectre I highlights consumer concern over privacy in an age of pervasive surveillance, underscoring the need for effective privacy protection technologies.
- Deveillance has announced the Spectre I, a portable microphone jammer aimed at blocking recording devices and detecting nearby microphones.
- The Spectre I is set for release in the second half of 2026 with a price of $1,199.
- Critics express skepticism about the technology's feasibility, citing challenges in physics and the effectiveness of detecting microphones.
- The device's AI is designed to produce cancellation signals to obstruct automatic speech recognition technology, but concerns about noise levels persist.
Deveillance is a startup founded by a Harvard graduate, developing technologies to enhance consumer privacy in response to the rise of always-listening AI devices. The Spectre I's features intend to address the growing anxiety surrounding surveillance and privacy violations. Further development and studies are required to validate its claims.
Read the original ↗
Best AI Project Management Tools
As businesses increasingly depend on efficient project management, understanding the best AI tools available can significantly enhance productivity and streamline workflows.
- The article reviews several AI-powered project management tools including ClickUp, Asana, Monday.com, Motion, and Wrike, emphasizing their unique features and benefits.
- ClickUp integrates various project management methodologies and uses AI for task automation and risk prediction, significantly improving team coordination.
- Asana offers work orchestration and integrates with major platforms like Slack and Google Drive, focusing on team collaboration and task management.
- Monday.com provides a dynamic system for sales and operations, leveraging AI for automation and data visualization, though it has accessibility concerns.
- Motion and Wrike cater to specific needs like scheduling and enterprise management, respectively, each with distinct advantages and complexities.
AI project management tools assist teams in organizing tasks and improving efficiencies in workflows. The rise of remote work has spurred demand for such technologies, which can adapt to diverse project needs. Users should look for tools that meet specific project complexities and team dynamics.
Read the original ↗
The Future of Iran’s Internet Is More Uncertain Than Ever
The ongoing internet blackout in Iran underscores the regime's increasing control over information and highlights the challenges faced by citizens in accessing global communications amid a humanitarian crisis.
- Over 90 million Iranians have experienced a total internet blackout for more than six days, following heightened military actions against the country.
- Connectivity has dropped by 99 percent, with remaining access largely restricted to government officials and elites.
- Civil society groups have developed circumvention tools like Conduit to access the global internet, though these tools are severely hampered during total shutdowns.
- Surveillance practices continue as the government promotes the National Information Network, an internal system designed to control citizen access to information.
- Destruction of critical internet infrastructure due to air strikes poses additional challenges to connectivity recovery, even if the shutdown is lifted.
The Iranian government has repeatedly implemented internet shutdowns to suppress dissent, particularly during anti-government protests. Many citizens have resorted to VPNs and other bypass tools, but ongoing conflicts have complicated their efforts. The future of internet access in Iran remains uncertain amid increasing government censorship and military operations.
Read the original ↗
SkillNet: Create, Evaluate, and Connect AI Skills
SkillNet's approach addresses a major limitation in AI development by providing a structured way to improve agent performance through the systematic accumulation of skills.
- SkillNet introduces an open infrastructure for creating, evaluating, and organizing AI skills using a unified ontology.
- It integrates a repository of over 200,000 skills and a Python toolkit for diverse AI applications.
- Experimental evaluations indicate that SkillNet enhances agent performance, with average rewards improving by 40% and execution steps reducing by 30% across different models.
SkillNet aims to resolve the inefficiencies in AI skill transfer, allowing agents to build upon previous knowledge rather than start from scratch. The framework promotes lifelong learning and richer skill development in AI. Future applications may further refine AI capabilities in complex tasks and interactions.
Read the original ↗
OpenAI Introduces Codex Security in Research Preview for Context-Aware Vulnerability Detection,...
The introduction of Codex Security by OpenAI could enhance the security of software development by improving vulnerability detection and patch generation across various codebases.
- OpenAI has introduced Codex Security in a research preview for context-aware vulnerability detection.
- The tool focuses on validating and generating patches for vulnerabilities across codebases.
OpenAI is a prominent artificial intelligence research organization known for its advancements in AI technologies. This development follows a growing need for security measures in software development, as vulnerabilities can lead to significant risks. Future updates will likely reveal more about its effectiveness in real-world applications.
Based on the headline only — the full article could not be fetched.
Read the original ↗
When AI Companies Go to War, Safety Gets Left Behind
The escalating AI arms race between companies and military entities poses significant risks to global safety and oversight, undermining previous commitments to AI regulation and responsible deployment.
- The Pentagon is pressuring Anthropic to remove limitations on the use of its AI models for military applications, leading to contract termination.
- Anthropic's Responsible Scaling Policy, aimed at mitigating AI risks, has fallen short amid rising competition and a lack of federal regulation.
- The competition among AI companies has intensified, pushing them to prioritize speed and contracts over safety measures.
AI companies, previously united in calls for safety and regulation, now face an environment characterized by competition and military pressure. Anthropic's dealings with the Pentagon highlight a larger concern for unregulated AI development in warfare. Observers should monitor how this landscape evolves amid increasing calls for AI oversight.
Read the original ↗