Articles from Source: The-New-Stack

Fable is coming back: Federal government lifts export controls on Anthropic AI model

2026-07-01 00:08
📢 Exciting news for AI enthusiasts! The U.S. Commerce Department has lifted export controls on Anthropic's AI models, Fable 5 and Mythos 5. This allows Anthropic to restore access to Fable for users after it was temporarily suspended due to security concerns. 🔓 In a letter to Anthropic, Commerce Secretary Howard Lutnick highlighted the company's commitment to address security risks and inform the government of any malicious activity. Stay tuned for updates from Anthropic as they bring Fable...
Source: The New Stack
Frederic Lardinois

Anthropic’s Claude Sonnet 5 system card says more about the future of AI than its benchmarks do

2026-06-30 20:37
🚀 Anthropic's Claude Sonnet 5 has launched, showcasing advancements in coding and reasoning. 📊 While benchmark results are highlighted, the 145-page system card emphasizes how AI agents operate autonomously and handle challenges like prompt injection and recovery from errors. 🔍 Key evaluations include malicious use and stealth behavior, indicating a shift towards reliable AI performance in real-world applications. 🛠️ Organizations will need new infrastructure to support these advanced agents...
Source: The New Stack
Amanda Caswell

The infrastructure lock-in costing AI companies hundreds of millions

2026-06-30 19:04
The AI infrastructure landscape is shifting, as highlighted by Jim Keller, CEO of Tenstorrent. He emphasizes that optimizing for current models may lead to costly lock-in as needs evolve. AI is outgrowing the infrastructure built for it, requiring a rethink of hardware strategies. It's no longer just about the fastest GPU; balance and adaptability are key. Nvidia's new Vera Rubin platform aims to address these challenges with a multi-chip system designed for diverse workloads. #AI...
Source: The New Stack
Amanda Caswell

Anthropic Sonnet 5: It closes the gap with Opus 4.8, and is cheap until August

2026-06-30 18:00
🚀 Anthropic has launched Sonnet 5, claiming it as their most advanced Sonnet model yet. It closely matches the performance of Opus 4.8, particularly excelling in reasoning, tool use, and coding tasks. While it doesn't surpass Opus 4.8, it offers a more affordable option until August. 💰 Introductory pricing is set at $2 per million input tokens and $10 for output, encouraging developers to test the model. #Anthropic #Sonnet5 #AI #MachineLearning #TechNews
Source: The New Stack
Frederic Lardinois

Anthropic launches Claude Science, an AI workbench for scientific research

2026-06-30 17:00
🔬 Anthropic has launched Claude Science, a new AI workbench designed for scientific research. This application runs on macOS and Linux, allowing researchers to conduct their work in one centralized place. Currently in beta, Claude Science primarily targets life sciences but is expected to expand in the future. 🛠️ Users on Claude Pro, Max, Team, and Enterprise plans can access it, with discounted options available for research labs. The tool aims to streamline research by integrating multiple...
Source: The New Stack
Frederic Lardinois

AWS launches a desktop for agents

2026-06-30 16:01
🚀 AWS has officially launched Amazon WorkSpaces for Agents, following a brief public preview. These cloud-based desktops allow agents to use desktop applications without needing custom integrations. Wolters Kluwer, a key customer, highlights the benefits of integrating AI into their workflows. The service uses the Model Context Protocol (MCP) for efficient session management and allows human oversight for better control. #AWS #CloudComputing #WorkSpaces #AI #VirtualDesktops
Source: The New Stack
Frederic Lardinois

AWS just put $1 billion into forward deployed engineers. Here’s why it matters for enterprise teams.

2026-06-30 15:02
🚀 Amazon Web Services is investing $1 billion in a new initiative called forward deployed engineering. This organization will place AWS engineers directly with enterprise teams to assist in building AI applications. 🤝 The goal is to provide hands-on support as AI projects transition into production. AWS has noted that customers prefer engineers embedded within their teams for better collaboration and expertise development. 🔗 These teams are already collaborating with various organizations,...
Source: The New Stack
Amanda Caswell

“The harness is where the hard work is”: Harness bets on agents that enterprises can trust in production

2026-06-30 15:00
Harness has launched Autonomous Worker Agents, a new feature that allows enterprises to replace fixed scripts in delivery pipelines with AI agents. 🤖 These agents can handle tasks such as deployment, testing, and security scans while adhering to existing governance controls. This enhances flexibility without compromising safety. 🔒 CEO Jyoti Bansal emphasizes that managing agents in production presents unique challenges compared to coding agents. The new agent-file format simplifies the...
Source: The New Stack
Frederic Lardinois

Aikido acquires Root to backport open source fixes without forcing upgrades

2026-06-30 14:00
Aikido Security has acquired Root for $70 million, enhancing its cybersecurity offerings. This acquisition integrates Root’s technology into a new product called Aikido Libraries, allowing for backporting security fixes to existing open source packages without requiring upgrades. 🔒 Aikido commits to providing these fixes for critical vulnerabilities listed in CISA's catalog at no cost, benefiting the entire open source community. 🌐 Founded in 2022, Aikido continues to expand its capabilities...
Source: The New Stack
Paul Sawers

Operating Kubernetes at scale: a few stories from running Amazon EKS

2026-06-30 13:00
Amazon EKS operates hundreds of thousands of Kubernetes clusters across 30+ AWS regions. 🛠️ Key insights reveal that many availability issues arise not from component failures, but from how components respond to problems. A resilient control plane prevents faults from escalating into outages. As workloads evolve, including AI and analytics, EKS adapts its systems to handle rapid state changes and higher demands. This evolution ensures that EKS supports mission-critical workloads effectively....
Source: The New Stack
Neelendra Bhandari

IdentityServer4 is dead. Here’s what comes next.

2026-06-30 08:00
🚨 Big changes in identity management! Duende's commercialization of IdentityServer4 in December 2022 left many developers dissatisfied. In response, Rock Solid Knowledge (RSK) has forked the project to create Open.IdentityServer, an open-source alternative. This new platform aims to provide a free, supported path for developers without the pressure of immediate monetization. RSK emphasizes that open-source success comes from adoption first. The initial release of Open.IdentityServer was on...
Source: The New Stack
Adrian Bridgwater

Base44 bets a narrow model beats frontier AI for vibe coding

2026-06-29 18:03
🚀 Base44 has introduced its proprietary AI model, Base One, designed for vibe coding web applications. This model is a fine-tuned version of an existing open-source LLM, marking a notable first in app creation platforms. Founder Maor Shlomo highlights its reinforcement learning approach, enhancing the model using real platform tasks. Base44 aims to optimize application development, reducing reliance on general AI models and external vendors. The company has seen remarkable growth, with...
Source: The New Stack
Darryl K. Taft

JetBrains kills Kotlin Notebook months after Microsoft’s Polyglot exit. But Jupyter is doing just fine.

2026-06-29 17:59
JetBrains has announced that it will sunset Kotlin Notebook, an interactive coding plugin launched in July 2023. Starting with IntelliJ IDEA 2026.2, the plugin will be handed to the open-source community under the Apache 2.0 license. The decision follows low adoption rates, with JetBrains acknowledging that AI tools have shifted developer workflows. This move mirrors Microsoft's earlier decision to deprecate Polyglot Notebooks for similar reasons. The Jupyter format remains popular, as it has...
Source: The New Stack
Paul Sawers

Palantir and Nvidia want to change who owns government AI

2026-06-29 17:55
Palantir and Nvidia are shifting the landscape of AI ownership in government sectors. Their recent collaboration focuses on an "intelligent engine" that enables organizations to run and customize AI within secure environments, keeping data in-house. This approach allows for better governance and compliance without relying on external cloud services. The partnership leverages Nvidia's Nemotron models, which are designed for efficient operation on internal GPUs, promoting a shift from using...
Source: The New Stack
Amanda Caswell

“Bring it to our shop”: Workday’s pitch for keeping AI agents close to your most valuable data

2026-06-28 15:00
Workday is advancing its AI capabilities, focusing on accuracy for payroll and HR data. CTO Gabe Monroy emphasizes that even a 99% success rate is insufficient due to the critical nature of this data. At the recent DevCon conference, Workday introduced Agent-Ready Tools to enhance agent reliability, including the Model Context Protocol and Agent Passport for testing and monitoring agents. Monroy, with a background in AI infrastructure, believes that safety in AI should be a foundational...
Source: The New Stack
Frederic Lardinois

Okta is the first to bring AI agent governance inside FedRAMP boundaries

2026-06-28 13:49
🚀 Okta has launched its AI agent governance platform for FedRAMP and HIPAA-regulated environments, marking a significant step in identity management. This platform treats AI agents as first-class identities, enhancing security and compliance for federal agencies and healthcare organizations. The product addresses risks associated with unmanaged agents, including compliance violations and increased breach potential. With a focus on governance, it ensures agents are registered and monitored...
Source: The New Stack
Darryl K. Taft

Greptile, Cursor, and Devin agree that agents should run their code. What they run it against matters.

2026-06-27 15:00
The tech industry is rapidly moving towards runtime verification for agent code. 🚀 Companies like Greptile, Cursor, and Devin emphasize that agents should not only write code but also verify it in real-time. Static checks alone are insufficient. By allowing agents to run their code before human review, efficiency increases and bottlenecks are reduced. 🔄 However, challenges remain, especially in ensuring that the runtime environment accurately reflects real system interactions. As the trend...
Source: The New Stack
Arjun Iyer

Vibe slop is the symptom. Context debt is the disease.

2026-06-27 10:30
Engineers Armin Ronacher and Mario Zechner highlight a growing issue in coding: "vibe slop." This term describes inefficient and error-prone code generated through prompt-based tools. They warn that this can lead to increased compute and resource costs, affecting startups' sustainability. Additionally, Postman's CEO Abhinav Asthana introduces the concept of "context debt," which refers to the complex web of services and APIs that complicate development. This debt accumulates over time and...
Source: The New Stack
Matt Burns

After Fable 5 ban, Anthropic and 19 organizations launch open source security body

2026-06-26 18:36
🚀 The tech industry is uniting to enhance open-source security with the launch of Akrites, led by the Linux Foundation and backed by 19 organizations, including AWS, Google, and Microsoft. 🛡️ This initiative aims to streamline the discovery, remediation, and disclosure of vulnerabilities in critical open-source software. AI models are now making it possible to identify multiple vulnerabilities quickly, but attackers have access to these tools as well. 🔍 As AI tools reveal thousands of...
Source: The New Stack
Paul Sawers

The US government just told OpenAI who’s allowed to use the next GPT 5.6 model

2026-06-26 18:26
🚨 The US government has mandated restrictions on OpenAI's upcoming GPT 5.6 model due to cybersecurity concerns. 🤖 Only a select group of government-approved partners will have access during the initial release. This follows a similar directive given to Anthropic for their AI models. 🔍 OpenAI CEO Sam Altman confirmed that access will be granted customer by customer, which is an unusual approach for such technology. #OpenAI #GPT56 #AI #Cybersecurity #Innovation
Source: The New Stack
Adrian Bridgwater

AWS, Microsoft, and Google agree the session is the new unit of compute. They disagree on how to isolate it.

2026-06-26 14:11
Four AI leaders—AWS, Microsoft, Google, and Anthropic—are advancing towards a common architectural shift in agent runtime updates. 🤖 While they agree that the session is the new unit of compute, they differ on how to achieve isolation. Microsoft focuses on per-session isolation, while AWS uses microVMs and Google applies dedicated sandboxes. This evolution highlights the need for session-aware execution in enterprise agents, moving beyond traditional request-level load balancing. 🌐 #AI...
Source: The New Stack
Janakiram MSV

Your engineering org needs an AI slop registry

2026-06-26 14:00
AI coding tools can accelerate coding but also amplify mistakes across projects. These errors often include subtle misalignments with project needs, leading to bloated or incorrect code that can pass initial checks. Teams typically try to improve AI instructions with files like CLAUDE.md, but this doesn't always ensure mistake detection. An independent verification system is essential for catching these errors effectively. Aviator's approach prioritizes verified intent over traditional code...
Source: The New Stack
Ankit Jain

Shopify taught AI to spot duplicate products. Here’s why retailers are scrambling.

2026-06-26 13:52
Retailers are adapting to a shift in shopping trends as AI shopping agents become more prevalent. Shopify's new system, Catalog, helps AI identify and group duplicate product listings, enhancing product discoverability. This comes as AI-driven traffic to Shopify stores has surged significantly. As the landscape evolves, retailers are rethinking how they present product information to AI. 📈🤖 #RetailTech #AI #Shopify #Ecommerce #ProductDiscovery
Source: The New Stack
Suman Bhattacharyya

The AI agent identity problem nobody’s talking about

2026-06-26 13:00
Many AI agent projects encounter challenges during security reviews due to unclear identity models and broad permissions. This issue can halt progress when questions arise about the identity under which agents operate. Research indicates a significant increase in AI agents, highlighting the growing need for well-defined identities to ensure accountability and security. To address this, developers must focus on key identity decisions, such as using workload identities instead of shared service...
Source: The New Stack
Jackson Connell

Public cloud vs. on-prem: Summit on where each workload belongs

2026-06-26 00:09
The debate on public cloud vs. on-prem solutions is heating up. As AWS and other cloud services grow, companies are reassessing their cloud strategies due to rising costs. Many now realize that not all workloads are best suited for the cloud. Experts suggest that segmenting workloads and moving some back on-prem can lead to lower costs and better management of high-risk data. For insights on the evolving landscape, check out the latest episode of The New Stack podcast. 💻☁️📈💡 #CloudComputing...
Source: The New Stack
Alex Wilhelm

“Code should be regenerated, not maintained”: Codeplain makes the case for spec-driven development

2026-06-25 17:47
AI is rapidly generating code, which presents new challenges for developers. A growing number advocate for a shift from code maintenance to spec-driven development. Dušan Omerčević, CEO of Codeplain, emphasizes that specifications should be the focus, allowing for easier updates and regeneration of code rather than maintaining it. Codeplain recently introduced an open-source framework called plain-forge to streamline the development process by automating spec creation. They’ve also secured...
Source: The New Stack
Paul Sawers

Trust in Rust: Foundation debuts official training to tackle steep learning curve

2026-06-25 16:30
🚀 The Rust Foundation has launched the Rust Foundation Trusted Training (RFTT) program, a formal accreditation for Rust training providers. This initiative aims to ensure quality and build trust in Rust training, as adoption grows across various industries. Five organizations are part of the founding cohort. 🏢 Accreditation evaluates five key areas, ensuring a high standard of training. Fees for accreditation cover administrative costs and are set to support the program's operations. For more...
Source: The New Stack
Darryl K. Taft

Template-based data extraction is dead. Here’s what comes next.

2026-06-25 16:00
Businesses are facing challenges in managing unstructured data like PDFs and audio recordings. Traditional template-based extraction methods are becoming outdated due to the diverse formats of modern documents. 📄🔍 Amazon Bedrock Data Automation (BDA) offers a solution. This generative AI-powered service automates the extraction and transformation of various content types, ensuring faster and more accurate processing. BDA uses Foundation models for intelligent content understanding and allows...
Source: The New Stack
Hafiz Hassan

How we cut AI costs by 80%

2026-06-25 15:00
AI costs are increasing, but not just due to more expensive models. A major factor is messy context management in enterprises. 🤖 Context accumulation drives costs as agents repeatedly gather information from multiple sources like GitHub and Jira. This results in high token consumption for queries. 📊 To address this, an experiment was conducted to streamline context retrieval by creating a semantic layer that reduces unnecessary data hops. The goal is to lower costs while maintaining...
Source: The New Stack
Zohar Einy

Agent Toolkit for AWS includes 20+ agent skills, but your agent might not load them without this one file

2026-06-25 14:00
🚀 The Agent Toolkit for AWS has launched, featuring over 20 agent skills designed to enhance your workflow. However, to fully utilize these skills, one crucial file is necessary: the rules file. This file instructs your agent to access skills and resources effectively. Setting up takes just two minutes. Make sure to check the README for complete instructions. 📄💻 #AWSToolkit #AgentSkills #TechTips #CloudComputing #MCP
Source: The New Stack
Esin Saribudak

OpenAI wants to claim more of the AI stack with Jalapeño, its first custom chip

2026-06-24 20:55
OpenAI has unveiled Jalapeño, its first custom inference accelerator, developed with Broadcom and supported by Celestica. This chip aims to enhance the performance of large language models (LLMs) while making AI faster and more cost-effective. OpenAI is positioning itself to control a greater share of the AI stack. The move reflects a trend among tech giants, as companies like Google and Amazon have also created their own AI chips. 🔍💻🌐 #OpenAI #Jalapeño #AI #TechNews #Innovation
Source: The New Stack
Meredith Shubel

Azul wants to find your unpatched JVMs before AI does

2026-06-24 20:53
🔍 Azul Systems is offering a free vulnerability risk assessment for Java Virtual Machines (JVMs). This tool helps DevOps and SecOps teams identify unpatched JVMs and enhance security before AI-driven threats emerge. The assessment scans networks to find JVM instances and provides a prioritized remediation roadmap, referencing established vulnerability databases. Azul's approach targets its security-only Critical Patch Updates, aiming to minimize risks during Java updates. They claim this...
Source: The New Stack
Darryl K. Taft

Will it Mythos? One coder’s verdict on Anthropic’s blend of debugging

2026-06-24 20:51
Anthropic's journey with Claude Fable 5 and Mythos 5 has sparked interest among developers. Joe Cooper, a Texas-based coder, expresses skepticism about Mythos's ability to find complex security bugs. He aims to create a benchmarking service to test Mythos against other models, using his tool, Nelson, for bug hunting. Cooper’s analysis, titled "Will It Mythos?", draws inspiration from a popular YouTube series. He highlights the differences in bug identification across models and mentions that...
Source: The New Stack
Adrian Bridgwater

Chainguard targets Java’s unpatched vulnerability backlog with drop-in remediated libraries

2026-06-24 20:48
Chainguard has launched its Libraries for Java, addressing unpatched vulnerabilities in legacy Java systems. This solution focuses on the Spring Boot ecosystem, offering backported fixes for critical CVEs across multiple frameworks. With the rise of AI-assisted scanning tools, the security landscape has become more challenging, generating numerous vulnerability reports each month. Many Fortune 500 companies still rely on outdated frameworks, increasing their risk. Chainguard's approach allows...
Source: The New Stack
Darryl K. Taft

Agentic infrastructure operations begin with accurate, reliable infrastructure data

2026-06-24 16:37
Organizations are increasingly leveraging AI for infrastructure management, targeting automation and efficiency. 🤖 However, the effectiveness of these AI agents hinges on the quality of infrastructure data. Inaccurate data can lead to serious issues, such as data exposure and system outages. 📉 Many enterprises struggle with fragmented tools, which create gaps in their infrastructure visibility. A unified model of infrastructure is essential for reliable AI operations. Infrastructure...
Source: The New Stack
Doug Sillars

Sakana Fugu is more than a router. But it’s not the blueprint for AI sovereignty, either.

2026-06-24 12:30
Sakana AI recently launched Fugu, a multi-agent orchestration system aimed at enhancing performance while minimizing reliance on single providers. 🤖 Fugu reportedly matches benchmarks set by competitors, breaking tasks into subtasks and routing them across expert agents. However, user feedback is mixed regarding its performance and cost. While positioned as a step towards AI sovereignty, some experts express skepticism about its effectiveness. Fugu's internal logic is based on proprietary...
Source: The New Stack
Meredith Shubel

Kubernetes teams trust automation to ship code but not to touch CPU, and AI is raising the stakes

2026-06-23 20:56
Kubernetes teams rely on automation for deployments but hesitate to automate CPU and memory adjustments. 🤔 A recent survey of 321 practitioners shows that 82% trust automated delivery, yet 71% require human review for resource optimization. The fear stems from the risks involved in altering resource allocations, which can affect application stability. ⚙️ As AI workloads become more common, the need for efficient resource management grows. The cost of manual oversight is increasing, especially...
Source: The New Stack
Yasmin Rajabi

Anthropic gives @Claude a permanent seat in your Slack channels

2026-06-23 18:50
🚀 Anthropic has introduced Claude Tag, a new feature that embeds Claude into Slack as a persistent team member. This allows Claude to accumulate knowledge and work asynchronously without constant prompts. Available in beta for Enterprise and Team plan customers, Claude Tag replaces the previous Claude integration. It enhances collaboration by enabling Claude to monitor discussions, flag important information, and follow up on unresolved threads. Each channel has its own Claude, promoting...
Source: The New Stack
Paul Sawers

OpenClaw and Hermes agree on what an agent is. They disagree on what controls it.

2026-06-23 17:56
At Microsoft’s Build keynote, CEO Satya Nadella highlighted a shift to agentic AI, showcased by OpenClaw, an open-source project enabling continuous operation of AI agents. OpenClaw connects agents to existing channels, while Hermes Agent focuses on memory for skill enhancement. Both projects aim to refine agent controls, emphasizing different core components. This evolving landscape in AI harness technology is crucial for future developments. 🤖🔗 #AI #OpenSource #TechInnovation #Microsoft...
Source: The New Stack
Janakiram MSV

Developers are now validating code they didn’t write — and may not understand

2026-06-23 17:05
🚀 GitLab's recent AI Accountability Report highlights the challenges developers face with AI-generated code. 91% of organizations now use multiple AI coding tools, boosting coding speed. However, 43% struggle to differentiate between AI and human-written code, raising concerns about control over software quality. Manav Khurana emphasizes the shift in the bottleneck from writing to reviewing code, with 85% of respondents noting increased validation workloads. To address these issues, he...
Source: The New Stack
Adrian Bridgwater