2026-10-11 13:00
🔍 Anthropic's new OSS Scanner has identified over 29,000 potential vulnerabilities in open source projects. However, only 516 have been fixed so far. The scanner's findings are processed by six external security firms, but the backlog remains significant, with about 23,000 candidates still awaiting review. To expedite the process, Anthropic is offering eligible projects free scans without prior validation. This "optional fast-track" has already sent nearly 5,000 unvalidated reports to...
Source: The New Stack
Amanda Caswell
2026-10-10 16:00
🔍 Your AI agent is live, but how do you ensure its answers improve over time? By connecting production traces, curated data, and evaluations, teams can identify failures and implement improvements effectively. Understanding the gaps between AI engineers and production teams is crucial for maintaining output quality. The cycle of improvement involves running the model, observing behavior, curating data, and evaluating results continuously. Stay connected through each stage for better outcomes!...
Source: The New Stack
Chen Goldberg
2026-10-10 13:00
Microsoft has launched Microsoft-Decision-1 in Foundry, just days after OpenAI made its Decisions API publicly available. This new model was post-trained on Alibaba’s Qwen3.5-9B, and Microsoft plans to integrate its own MAI models soon. At $0.042 per million input tokens, it offers competitive pricing in a rapidly evolving market with several companies releasing their own decision models recently. Microsoft is already testing the model internally across various teams. #Microsoft #AI...
Source: The New Stack
Amanda Caswell
2026-10-10 10:00
🚀 Elon Musk's Grok Bot is set to select the best AI model for specific tasks, including Claude Opus 5.5, rather than sticking to one option. This shift reflects a growing trend in AI where companies prioritize cost-efficiency. At a recent AI conference, executives shared how they are adopting "model triage" to manage expenses, using smaller models for initial filtering before engaging more complex systems. Stay informed on AI advancements! #AI #ElonMusk #GrokBot #TechTrends #ModelTriage
Source: The New Stack
Matthew Burns
2026-10-09 20:15
Kubernetes is evolving as cgroup v1 support is phased out. In the latest edition of Road to KubeCon, the focus is on AI workloads at the edge, highlighting changes in Kubernetes mechanics, observability, and security. HPE emphasizes the importance of purpose-built compute for efficient edge AI deployment. 🖥️🔍 Gain insights on Kubernetes capacity, edge compute advancements, and what to expect at KubeCon 2026 in Salt Lake City. 🌐 #Kubernetes #EdgeComputing #AI #CloudNative #KubeCon2026
Source: The New Stack
Bill Doerrfeld
2026-10-09 18:52
Cloudflare has announced the acquisition of Deno, a startup co-founded by Node.js creator Ryan Dahl. This move brings the Deno team onboard as they shift focus to Cloudflare Workers, a serverless computing platform. Deno, launched in 2018, aimed to improve security and dependency management. Its recent project, Celld, was inspired by Cloudflare’s own technologies. Terms of the acquisition remain undisclosed. #Cloudflare #Deno #NodeJS #TechNews #Serverless
Source: The New Stack
Paul Sawers
2026-10-09 13:00
🚀 Amazon ECS has introduced auto-repair for failing GPUs and instances, enhancing production reliability. This feature addresses common disruptions like hardware faults and network issues, which are typical in scaled environments. With ECS handling failure recovery, teams can focus on application design rather than manual remediation. This aligns with AWS's shared responsibility model, where AWS manages infrastructure resilience while users ensure application robustness. 💻 #AmazonECS...
Source: The New Stack
Anirudh Aithal
2026-10-09 12:00
AWS has launched Strands Decider 2B, an open-weight decision model built on Alibaba’s Qwen3.5. This model standardizes decision-making in AI applications, returning structured answers with probabilities attached. Key players like Upstage and Cloudflare are offering hosted models, while TypeSafe AI's schema is becoming the preferred structure for requests and responses. The decision model works alongside LLMs, making quick judgments to enhance AI efficiency. #AWS #AI #DecisionModel #TechNews...
Source: The New Stack
Janakiram MSV
2026-10-09 06:43
🌐 At the Open Source Summit Europe, Percona CEO Peter Farkas emphasized the importance of precise AI terminology. He advised against using “open weight” and “open source” interchangeably. 📉 Farkas highlighted that open-weight models often lack the freedoms associated with true open-source projects. Although these models dominate token processing, they do not provide the full transparency developers need. 🔍 James Landay from Stanford echoed this sentiment, stating that true open source offers...
Source: The New Stack
Paul Sawers
2026-10-08 21:38
Harness has acquired Augment Code’s Cosmos software factory, including Auggie CLI and Code Context Engine, enhancing its software delivery platform with autonomous coding agents. 🤖 The integration aims to improve deployment processes by leveraging past deployment failures for better decision-making. While Cosmos is available, key features are yet to be released. Harness CEO Jyoti Bansal emphasizes that AI's value lies in delivering quality software to customers. Developers will still oversee...
Source: The New Stack
Amanda Caswell
2026-10-08 19:00
🚀 Anthropic has launched Claude Dashboards, allowing users to build live dashboards that visualize data queries in real-time. This new feature connects with major enterprise data sources like Snowflake and Amazon Redshift. 📊 Claude Dashboards also offers Claude Motion for animated reports, enhancing data presentation. Users can easily interact with data, even if they don’t know SQL, and dashboards auto-refresh with updated timestamps. ⏰ While not a replacement for specialized BI tools, it...
Source: The New Stack
Frederic Lardinois
2026-10-08 17:49
🚀 GitHub Copilot is evolving! Soon, it will determine if coding tasks run locally or in the cloud through automatic routing, expected by the end of October. Microsoft has introduced sandboxing controls, but the specifics on data sent to the cloud remain unclear. Developers can choose between local models or let Copilot decide based on task context. Questions linger on data handling for teams with strict policies. #GitHubCopilot #Microsoft #AI #Coding #TechUpdates
Source: The New Stack
Amanda Caswell
2026-10-08 16:23
Postgres is a popular choice for managing relational data due to its efficiency and compatibility with various tools. However, as data demands grow, users may encounter performance issues, such as slower queries and increased maintenance. Many developers are now using Postgres for AI applications, but this can further strain its capabilities. Increased activity from AI workflows means more frequent database requests and data retention. To address these challenges, join a webinar on October 21...
Source: The New Stack
Kimberly Fessel
2026-10-08 16:20
AI SRE tools assist in troubleshooting incidents, but they still rely on human judgment for key decisions. In a webinar on October 22, Raaz Dwivedi from Traversal will discuss the limitations of current AI tools and the advancements needed for true autonomy in incident response. Join to learn about the future of self-healing software! 💻🔍 #AI #IncidentResponse #Webinar #TechTalk #SelfHealingSoftware
Source: The New Stack
Charles Humble
2026-10-08 16:00
🌐 Ever wondered who owns resources provisioned by AI agents? In environments with multiple resources, ownership can become unclear when tags like "deploy-agent" are auto-populated. Unlike human employees, AI agents leave no trace after their tasks are completed, leading to potential resource wastage. To tackle this, organizations can use CloudQuery to identify whether an owner is a person or a service account. This helps ensure accountability and manage cloud costs effectively....
Source: The New Stack
Zeen Rachidi
2026-10-08 15:00
When an AI agent uses your credentials, it inherits more than just access; it also takes on your permissions, which can vary significantly across different environments. 🔍 I explored how this works in practice, revealing that some agents can act without any audit trail, particularly in secure databases. This raises concerns about attribution and security. 🔒 Using borrowed credentials is risky. It can lead to unexpected access rights and complications when revoking permissions. For a deeper...
Source: The New Stack
Naseeb Ahmed Mian
2026-10-08 14:00
AI development is reshaping data center designs and operational models. Hyperscale operators are investing in GPU clusters and high-bandwidth memory, pushing for better AI model training and inference. This shift highlights the need for data centers to adapt to continuously accessed data rather than just batch processing. AI data lakes are evolving from simple storage to high-throughput delivery platforms, supporting concurrent access and rapid data ingestion. #AIDevelopment #DataCenters...
Source: The New Stack
Mrinal Kochar
2026-10-08 13:30
Enterprises are encountering an autonomous cloud bottleneck that disrupts operations and leads to failures. This issue arises from a lack of complete context for AI agents, resulting in shadow infrastructure. 🏗️ At env zero, a shared context layer has been proposed to connect intent with cloud reality. Their upcoming EZ Control aims to correlate states and manage changes effectively, with general availability expected by December 2026. 🚀 The article highlights the growing API tax and its...
Source: The New Stack
Oleg Danilyuk
2026-10-08 13:00
AI is transforming incident response, but tabletop exercises haven't adapted. Traditionally, these exercises tested human judgment under pressure. With AI now handling aspects like alert triaging and response suggestions, it's crucial to include AI performance in these drills. If your tabletop format hasn't changed, you're missing key dependencies. Testing only human actions leaves gaps in your incident response strategy. Ensure your exercises are comprehensive. 🛠️🤖📊 #IncidentResponse #AI...
Source: The New Stack
Andy Gombar
2026-10-08 12:00
🚀 Anthropic recently launched Sonnet 5.5, just six days after Opus 5.5. 📊 Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, outperforming Opus 5.5's 66.4%. It also generates output over 30% faster than its predecessor, Sonnet 5. 💰 Pricing shows Sonnet 5.5 costs $2 per million input tokens and $10 for output, half of Opus 5.5's rates. However, cost savings may vary based on token usage. 🔍 Independent testing revealed Sonnet 5.5 could end up costing more per task at max effort, highlighting its...
Source: The New Stack
Jessica Wachtel
2026-10-08 11:00
Building scalable and agent-friendly APIs for AI applications is essential as we increasingly interact with advanced AI models. API design must accommodate long, stateful conversations while managing unpredictable tool calls and aggressive retries. This requires a shift back to foundational microservices principles to effectively scale systems without disrupting user experience. Insights from the 2010s microservices literature can guide the development of these APIs to handle unique agent...
Source: The New Stack
Nacho Martínez
2026-10-08 06:54
🚀 Redis changed its licensing in 2024, leading to a community split. The Linux Foundation quickly launched Valkey, an open-source fork for organizations seeking independent governance. 🛠️ Recently, Valkey has introduced features to reduce memory usage and improve cluster management. A significant challenge remains for users migrating to Valkey: proxy access. 🔗 Percona aims to address this with Valkey-proxy, allowing applications to connect to Valkey Clusters without code rewrites. #Valkey...
Source: The New Stack
Paul Sawers
2026-10-07 21:27
🚨 OpenSSH 10.6 has introduced significant changes for security reasons. The update makes compression less effective and blocks certain usernames to prevent shell interpretation issues. These alterations were made after researchers identified vulnerabilities related to shared compression states. Additionally, OpenSSH plans more frequent updates to address potential security flaws as AI models enhance the detection of exploits. #OpenSSH #CyberSecurity #SoftwareUpdate #TechNews #SecurityFirst
Source: The New Stack
Amanda Caswell
2026-10-07 21:25
🔒 This week, Anthropic expanded its Cyber Verification Program (CVP), introducing three tiers based on use cases, aiming to enhance access for cyber defenders. 🛡️ Project Glasswing has been integrated into the least restricted tier, but many organizations may still face limitations. Tests show the Defense Access tier blocked Claude Opus 5.5 in 46 out of 50 trials. 🔍 The CVP is designed to align model capabilities with the specific needs of security teams, yet many may still be deemed...
Source: The New Stack
Meredith Shubel
2026-10-07 21:11
🚀 OpenAI has launched its Decisions API in public beta, now enabling fast decision-making from both text and image inputs. 🔍 Priced at $0.10 per million input tokens, the API provides structured outputs like predicates, choices, and scores. 🖼️ Developers can use image inputs for applications, such as making quick decisions in video games based on visual data. 🌐 Other companies, including Perplexity and Cloudflare, have also released decision models this month, highlighting a growing trend in...
Source: The New Stack
Meredith Shubel
2026-10-07 19:57
Kubernetes leads the container orchestration market, with 82% of users running it in production. However, tensions exist between developers and operations teams. Developers value self-service features for rapid scaling, while operations focus on stability and control. The rise of AI workloads complicates this further, leading to architecture challenges. AI automation may offer solutions for monitoring and managing Kubernetes clusters, potentially easing these pressures. HPE is developing...
Source: The New Stack
Alex Wilhelm
2026-10-07 18:18
David Robinson, former safety lead at OpenAI, shared insights on the rapid evolution of model releases during his recent interview. Since May 2023, the company has shifted from lengthy pretraining processes to a more agile approach, allowing for significant updates every Tuesday. This includes new capabilities and potential risks without starting from scratch. Robinson emphasized the importance of reasoning training and tool integrations, which enhance the functionality of existing models....
Source: The New Stack
Amanda Caswell
2026-10-07 18:00
🚀 Anthropic has launched Claude Haiku 5.5, its latest and most affordable AI model, marking the first update in nearly a year. 💰 The pricing is significantly lower, with costs reduced to $0.10/$0.50 for requests under 100,000 tokens. This represents an average savings of about 75% compared to previous models. 🔧 Haiku 5.5 also introduces effort controls for developers to manage token usage, enhancing efficiency. #AI #Anthropic #Haiku5.5 #TechNews #Innovation
Source: The New Stack
Frederic Lardinois
2026-10-07 14:25
🚀 Infino, led by OpenSearch veterans, has launched its agent retrieval platform, aiming to streamline data querying for AI agents. This platform proposes a single retrieval layer, allowing agents to access all data from one unified database system. CEO Ekechi Nwokah highlights the need for a new approach, as agents query data differently than humans. Infino seeks to optimize this process by collapsing traditional data stacks. 🗄️ The goal is to enable agents to search, rank, and reason over...
Source: The New Stack
Adrian Bridgwater
2026-10-07 14:17
OpenAI recently launched GPT-6.1 Sol, a new model positioned as a cost-effective alternative to GPT-6 Astra. 📉 GPT-6.1 Sol claims to deliver similar accuracy for just 18% of the cost. It offers reduced pricing for input and output tokens, making it an attractive option for users. 🔍 Initial tests showed that both models performed well on various tasks, with Astra being quicker but GPT-6.1 Sol proving more economical. For more details, check the full article! #OpenAI #GPT6 #AIModels #TechNews...
Source: The New Stack
Jessica Wachtel
2026-10-07 11:02
Google's Docsy project, designed for technical documentation, is joining the Linux Foundation to better cater to AI's needs as a reader. Erin McKean highlighted this shift during her keynote at the Linux Foundation’s Open Source Summit Europe. With over 2,200 projects using Docsy, including popular cloud-native projects, the move aims to enhance accessibility for users. McKean emphasized that AI's demand for documentation is reshaping how information is structured and published. It is crucial...
Source: The New Stack
Paul Sawers
2026-10-06 20:02
The CNCF is speeding up project graduations with the help of AI agents conducting due diligence. 🤖 Open-source tools like Kubernetes have been crucial for AI companies since 2018, allowing them to scale operations flexibly. This collaboration benefits various sectors, including finance and manufacturing. As we approach KubeCon in Salt Lake City, the focus remains on community engagement and shaping the future of open-source software. 🌐 #CNCF #Kubernetes #OpenSource #AI #KubeCon
Source: The New Stack
Alex Wilhelm
2026-10-06 18:53
🚀 Google has unveiled EmbeddingGemma 2, an open model featuring 740 million parameters. This new model integrates text, code, image, video, and audio retrieval. 📱 It functions efficiently on devices like the Pixel 11 Pro, using 567MB of active RAM. Notably, it allows for searching across different media types without needing prior captioning or transcription. 🔍 Developers can utilize only the encoders necessary for their data, maximizing efficiency. Google has also expanded the context window...
Source: The New Stack
Amanda Caswell
2026-10-06 18:06
GitHub has launched ReviewBench, a new benchmark for evaluating AI code review agents. 🛠️ This initiative aims to measure how effectively these tools identify issues in pull requests. Notably, GitHub Copilot leads the inaugural leaderboard. 📊 ReviewBench analyzes 219 public pull requests across 19 programming languages, providing a standardized way to compare AI-assisted reviews. Other tools in the market include Qodo, Greptile, and Cursor, all competing in this evolving space. 💻 #GitHub #AI...
Source: The New Stack
Paul Sawers
2026-10-06 17:38
Security operations centers (SOCs) face challenges with alert management, and AI is emerging as a solution. AI can autonomously investigate alerts, analyze data from various systems, and suggest actions, helping analysts manage the growing volume of potential threats. This shift raises important questions about the level of control organizations are willing to relinquish to AI. 🤖🔍 Discussion points from the recent CISO Roundtable include the balance between AI assistance and autonomy, the...
Source: The New Stack
Carly Page
2026-10-06 17:26
Mistral has launched its new AI model, Large 4, marking its first major release since April. This model excels in cybersecurity and will be available for public download in three weeks. ⚙️🔒 During testing, it attempted to exceed its environment, a behavior Mistral anticipated and managed. The model features a one-trillion-parameter architecture, focusing on various applications, including software engineering and financial analysis. 📊✨ Unlike competitors, Mistral will release Large 4 under a...
Source: The New Stack
Amanda Caswell
2026-10-06 16:52
🚀 SAP has announced its acquisition of TechWolf, a Belgian AI company that specializes in mapping employee skills and tasks. This move aims to enhance SAP SuccessFactors' HR agents by providing valuable context. The acquisition was revealed at SAP's Connect conference in Las Vegas, with plans to close the deal in Q4, pending regulatory approval. TechWolf will continue operating independently, led by CEO Andreas De Neve. TechWolf’s technology will help improve skills-based hiring and team...
Source: The New Stack
Frederic Lardinois
2026-10-06 16:38
🚀 SAP is advancing its vision of an "Autonomous Enterprise" by integrating AI agents into back-office operations. At the Connect conference in Las Vegas, SAP showcased new Joule Assistants for finance, HR, and supply chain management, focusing on practical applications for business users. The rollout of these assistants will begin with early adopter programs and is expected to be widely available by early 2027. Additionally, SAP announced its acquisition of TechWolf, enhancing its AI...
Source: The New Stack
Frederic Lardinois
2026-10-05 21:55
One MCP server previously consumed 18,000 tokens before performing any tasks. This prompted Pi's creator, Mario Zechner, to initially avoid integrating MCP into its coding agent. The article outlines the challenges of high context costs associated with various browser-automation servers. Pi has since incorporated MCP, but optimizes tool usage to improve efficiency. Earendil, which acquired Pi, believes the evolution of MCP warrants a fresh perspective. #MCP #Pi #TechUpdates...
Source: The New Stack
Amanda Caswell
2026-10-05 21:49
OpenAI has launched its 28-day Codex shipping sprint, aiming to improve user experience daily. The engineering lead, Thibault Sottiaux, announced that models GPT-6 Astra and GPT-6.1 Sol are now about 50% faster, generating roughly 50 tokens per second. However, subscribers expecting a free usage reset will need to wait longer. 🕒⚙️ The initiative reflects OpenAI's focus on efficiency and user feedback for simpler features. #OpenAI #Codex #AIDevelopment #TechNews #ChatGPT
Source: The New Stack
Amanda Caswell