Must know

AI coding tools boost throughput but degrade code quality: A Carnegie Mellon University study of 807 GitHub repositories found that while AI coding assistants initially increased code generation by 30-40%, this productivity gain vanished after three months. More alarmingly, static analysis warnings rose by 30% and code complexity increased by over 40%, persisting long after the initial boost. This suggests that while AI can accelerate coding, it introduces significant technical debt and maintainability issues, a finding consistent with 2024 data from GitClear. Meanwhile, Salesforce deployed Cursor to 20,000 engineers, reporting double-digit gains in cycle time, quality, and throughput, with 90% adoption in months. This contradiction highlights the ongoing debate on AI’s true impact on code quality and developer experience. Read more
Critical vulnerabilities in Chainlit AI framework allow server exploitation: Two critical vulnerabilities (CVE-2026-22218, CVE-2026-22219) were discovered in Chainlit, a popular Python AI chatbot development framework with millions of downloads. These flaws allow unauthenticated attackers to read any file on the server and perform cross-site request forgery (CSRF) and server-side request forgery (SSRF) attacks without user interaction. This could lead to the exfiltration of API keys, sensitive credentials, and database contents, enabling lateral movement within enterprise networks. Users are urged to upgrade to Chainlit version 2.9.4 immediately to patch these issues. Read more
VS Code exploited by ‘Contagious Interview’ campaign: A new social engineering campaign, ‘Contagious Interview,’ is exploiting Visual Studio Code by embedding malicious code in tasks.json files within Git repositories. When developers clone and trust these repositories, the malicious code executes, deploying JavaScript payloads from remote servers (often legitimate platforms like Vercel) to establish persistent backdoors. This highlights a critical vulnerability in VS Code’s trust model, where merely opening and trusting a repository can lead to system compromise. Developers are advised to never trust unvetted repositories and to manually inspect configuration files like tasks.json and package.json before execution. Read more

Good to know

Amazon launches Bedrock AgentCore for AI agent deployment: Amazon introduced Bedrock AgentCore, a platform for deploying and operating AI agents with zero infrastructure management, offering modular services like Runtime, Memory, Gateway, Code Interpreter, and Observability, along with a CLI toolkit for rapid creation and integration with frameworks like LangGraph and CrewAI. Read more
LLVM bans unchecked AI-assisted code contributions: The LLVM compiler project has implemented a ‘human in the loop’ policy, banning AI-generated code contributions without thorough human review due to a surge in low-quality submissions, and explicitly forbidding AI tools for ‘good first issue’ tasks to preserve learning opportunities for beginners. Read more
Mesa 26.0-rc1 brings major GPU driver enhancements: Mesa 26.0-rc1 has been released, featuring significant improvements to AMD’s RADV Vulkan driver, Intel’s ANV driver, and NVIDIA’s open-source NVK driver, including new Vulkan API extensions and video encoding capabilities for RADV, with the stable release expected in February 2026. Read more
jQuery 4.0 drops legacy browser support, adds security features: jQuery 4.0, released after nearly a decade, removes support for Internet Explorer 10 and Microsoft Edge Legacy, integrates Trusted Types and Content Security Policy, and migrates to ECMAScript modules, resulting in a smaller, more modern library. Read more
Cloudflare fixes WAF bypass vulnerability: Cloudflare patched a vulnerability in its Web Application Firewall (WAF) ACME validation logic that allowed attackers to bypass security rules and access origin servers directly by exploiting improper validation of ACME challenge tokens. Read more
New Ruby framework CMDx uses CERO pattern: CMDx is a new Ruby framework that organizes business logic using composable command objects, implementing the CERO (Compose, Execute, React, Observe) architectural pattern with typed attributes, built-in validation, and structured logging for improved maintainability and traceability. Read more
bunq automates 97% of customer support with AI: Europe’s second-largest neobank, bunq, developed an AI customer support system called Finn using Amazon Bedrock and Anthropic’s Claude models, which now handles 97% of customer interactions, with 82% fully automated, achieving a 47-second average response time across 38 languages. Read more
OpenEvidence valued at $12B for AI medical platform: OpenEvidence, an AI-powered medical information platform for doctors, raised $250 million in Series D funding, reaching a $12 billion valuation, with its platform handling 18 million clinical consultations in December, competing directly with OpenAI’s ChatGPT Health and Anthropic’s Claude for Healthcare. Read more
NetBox 4.5 introduces object owner model for dynamic permissions: NetBox 4.5 now includes a native Owner model, allowing users and groups to be assigned as owners of infrastructure objects, which automatically links ownership to dynamic permission constraints, streamlining management and access control in multi-user environments. Read more
UniFi Fabrics launches for multi-site network orchestration: UniFi Fabrics is a new cloud orchestration platform for managing UniFi network infrastructure at scale, offering centralized configuration, security policy management, and API access for MSPs and IT teams, while maintaining a ‘local-first’ model to avoid vendor lock-in. Read more
Arctic Wolf observes Fortinet FortiGate SSO exploitation: Arctic Wolf detected an automated cyberattack campaign targeting Fortinet FortiGate devices by exploiting SSO accounts to make unauthorized firewall configuration changes, create persistence accounts, and exfiltrate data, closely resembling a December 2025 campaign exploiting SSO authentication bypass vulnerabilities. Read more
Ingram Micro ransomware breach exposes 42,000 records: Ingram Micro disclosed a July 2025 ransomware attack by the Safepay group that compromised over 42,000 current and former employee and applicant records, including names, contact details, and Social Security numbers, and paralyzed the IT distributor’s logistics operations for a week. Read more
OpenEvidence valued at $12B after Series D funding: OpenEvidence, an AI-powered medical information platform for doctors, raised $250 million in Series D funding, reaching a $12 billion valuation, with its platform handling 18 million clinical consultations in December, competing directly with OpenAI’s ChatGPT Health and Anthropic’s Claude for Healthcare. Read more
CUDA 13.1 simplifies GPU programming with new CUB API: CUDA 13.1 introduces a new single-call API for CUB (CUDA Core Compute Libraries) that simplifies GPU primitive algorithm usage by automatically handling memory allocation, eliminating the traditional two-phase pattern, and allowing custom execution options with zero performance overhead. Read more
PyTorch 2.10 expands AMD ROCm and Intel GPU support: PyTorch 2.10 has been released, bringing enhanced support and performance improvements for AMD ROCm and Intel GPUs, aiming to improve hardware compatibility within the deep learning ecosystem. Read more
LLM benchmarking reveals major API cost savings: Targeted benchmarking of LLMs on real customer support prompts, rather than generic leaderboards, can reduce API costs by 5-10x, as cheaper models often perform comparably to more expensive ones for specific use cases, according to an unnamed case study. Read more
Zanskar’s AI challenges geothermal energy estimates: Zanskar, a geothermal exploration startup, raised $115 million to use AI for identifying conventional geothermal resources, claiming a potential of 1 terawatt, significantly exceeding the U.S. Department of Energy’s estimate of 60 gigawatts by 2050. Read more
Reinforcement learning and LoRA advance LLM customization: Reinforcement learning (RL) is now central to enhancing LLM abilities, and when combined with Low-Rank Adaptation (LoRA), it enables rapid and compute-efficient experimentation, allowing for mass production and scaling of personalized AI agents, though experts note it won’t lead to AGI. Read more
Liquid AI releases LFM2.5-1.2B-Thinking for on-device reasoning: Liquid AI announced LFM2.5-1.2B-Thinking, a 1.2 billion parameter AI reasoning model designed to run fully on-device with less than 1 GB of memory, supporting a 32,768 token context window and explicitly generating ‘thinking traces’ for applications in agents, tool use, and RAG workflows. Read more
BlackIce toolkit targets AI security testing: BlackIce is an open-source toolkit bundling 14 distinct AI security testing tools into a single Docker image, simplifying deployment for prompt injection, jailbreak detection, data leakage, hallucination detection, and adversarial example creation, all mapped to MITRE ATLAS and DASF frameworks. Read more
OpenAI deploys age prediction for ChatGPT user safety: OpenAI is rolling out age prediction models for ChatGPT consumer accounts, analyzing behavioral signals to identify users under 18 and restrict access to sensitive topics, while also providing parents with new controls, though privacy advocates raise concerns about forced identity verification for incorrectly flagged users. Read more
OpenAI releases family-oriented AI literacy resources: OpenAI launched two new AI literacy resources for families, including a guide for teens on how AI models work and responsible usage, and a parent-focused resource with conversation starters, developed with feedback from online safety and teen development experts. Read more
Go Developer Survey 2025: satisfaction high, entry-level adoption declines: The Go Developer Survey 2025 reports 91% developer satisfaction with Go’s cohesiveness and tooling, but highlights challenges in identifying best practices and missing language features, noting a decline in entry-level adoption and mixed satisfaction with AI-powered tools due to code quality concerns. Read more
Microsoft releases security-fixed Go builds: Microsoft has released new builds of Go versions 1.25.6-1 and 1.24.12-1, both containing security fixes and available for download from Microsoft’s Go GitHub repository, without specifying the vulnerabilities addressed. Read more
OpenJDK January 2026 quarterly security update released: The January 2026 OpenJDK quarterly update, released on January 21, 2026, introduces 11 new security patches and hundreds of bug fixes across all supported Java versions (8, 11, 17, 21, and 25), with major distributors like Azul providing both Critical Patch Updates and Patch Set Updates. Read more
SmartOS platform overview and capabilities: SmartOS is a Type 1 hypervisor platform based on illumos, offering both OS virtualization (Zones/Containers) and hardware virtualization (KVM/Bhyve), running entirely from RAM and utilizing illumos technologies like ZFS, Crossbow, and DTrace for robust system management and security. Read more
Linux Foundation launches Agentic AI Foundation: The Linux Foundation announced the launch of its Agentic AI Foundation, focused on governance using MCP, alongside its January 2026 newsletter summarizing 2025 activities, releasing 2026 predictions, and highlighting new research on cyber resiliency and AI’s effects in Latin America. Read more
EU proposes unlimited telecom spectrum licenses: The European Commission’s Digital Networks Act proposes unlimited duration radio spectrum licenses for EU telecom operators to incentivize long-term investment, while explicitly rejecting requests for mandatory ‘Fair Share’ fees on Big Tech companies, opting instead for voluntary cooperation on infrastructure costs. Read more
Cybersecurity talent demand and IAM specialization trends: The cybersecurity sector is projected to grow 29% through 2034, with Identity and Access Management (IAM) roles being particularly lucrative and in demand, expanding to include non-human identities like AI agents, and requiring expertise in SCIM, SAML, and Zero Trust frameworks. Read more
Quantum networks face performance limits from selfish routing: Northwestern University researchers found that quantum communication networks can suffer from a ‘selfish routing’ paradox, where users optimizing for personal connection quality in decentralized networks can collectively degrade overall communication efficiency, challenging the belief that increasing entanglement always improves performance. Read more
KAUST launches National Quantum Hardware Foundry: King Abdullah University of Science and Technology (KAUST) launched the KAUST Quantum Foundry in Saudi Arabia, providing shared access to quantum cleanrooms and Process Design Kits (PDKs) to enable reproducible and scalable quantum hardware fabrication for superconducting, photonic, and hybrid quantum platforms. Read more
Alice & Bob proposes ‘Elevator Codes’ for quantum error correction: Alice & Bob proposed ‘Elevator Codes,’ a new quantum error-correction method designed to reduce logical error rates up to 10,000 times on cat qubit quantum computers by introducing active protection against bit-flip errors using a reusable logical ancilla qubit, potentially accelerating large-scale quantum computing timelines. Read more
Batmobile CUDA library accelerates equivariant GNNs: Batmobile, a custom CUDA kernel library, significantly accelerates equivariant graph neural networks (GNNs) by optimizing spherical harmonics and tensor product operations, achieving speedups of up to 20.8x on NVIDIA RTX 3090 hardware, which benefits molecular dynamics simulations and materials discovery workflows. Read more
Quantum machine learning applications in drug discovery: Quantum machine learning (QML) is moving from theory to practical use in drug discovery, with hybrid classical-quantum algorithms being applied to molecular modeling and optimization, though current demonstrations remain small-scale and depend on improvements in quantum hardware and error mitigation. Read more
MIT achieves faster cooling for trapped-ion quantum computers: MIT researchers developed a faster cooling method for trapped-ion quantum computers using integrated photonic chips with nanoscale antennas, achieving ion temperatures nearly 10 times lower than standard laser cooling in about 100 microseconds, making quantum computers more compact and efficient. Read more
Optical atomic clock innovation combines single-ion accuracy and multi-ion stability: Researchers at PTB and NIMT created a new optical atomic clock using ytterbium-173 ions, which combines single-ion clock accuracy with enhanced stability typically seen in multi-ion configurations, potentially advancing the redefinition of the SI second and offering applications in quantum computing and fundamental nuclear physics. Read more
Unified framework for spacetime fluctuation detection presented: Researchers at the University of Warwick introduced the first unified framework for detecting spacetime fluctuations, categorizing them into three types and translating theoretical predictions into measurable signals for laser interferometers like QUEST and GQuEST, extending beyond quantum gravity to dark matter and stochastic gravitational waves. Read more
Blue Origin reveals TeraWave satellite network: Blue Origin announced TeraWave, a new satellite internet initiative targeting enterprise, data center, and government sectors with a constellation of 5,280 LEO satellites delivering 144 Gbps via RF links and 128 MEO satellites supporting 6Tbps via optical links, with first deployment scheduled for late 2027. Read more
Google Cloud opens Bangkok region: Google Cloud launched a new regional data center in Bangkok, Thailand, as part of a $1 billion investment, aiming to provide low-latency connections, local data residency, and compliance with Thailand’s Personal Data Protection Act, while offering access to Google’s global AI capabilities like Gemini AI. Read more
Hetzner Cloud offers best performance-to-cost ratio: A comparison