Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
LLMs can be useful and thorough while still feeling untrustworthy because they speak with artificial confidence, invent facts, and imitate a grating human voice. Their behavior reflects the values and incentives of the companies that build them, which makes the discomfort more than a matter of style.
Frameworks still matter in AI-assisted development because they provide reusable abstractions, recognizable structure, and constraints that keep generated code coherent. Without them, coding agents tend to improvise unmaintained frameworks of their own, so the better response is to build stronger frameworks for both humans and agents.
Sponsored
AI works great for individuals but stalls for engineering enterprises. In this new O'Reilly book (early release compliments of Antithesis [14]), Peter Bell provides a hands-on guide to the real challenges of AI adoption: aligning teams, evolving org structures, adapting hiring strategies, and managing risk. Claim your free copy
Grok 4.7 uses a larger base model and longer reinforcement learning to improve coding, long-running work, self-verification, and professional document tasks.
Gortex indexes code into a local graph and exposes cross-repository search, references, call chains, and impact analysis through a CLI, MCP server, API, and web interface. It supports 257 languages and claims up to 50 times fewer tokens per response than naive file reads by returning only the relevant graph context.
Kev is a family of small, open-source decision models built on Qwen3.5 that can run locally and expose a TypeSafe System One-compatible API. The repository includes 0.8B, 4B, and 9B checkpoints, training and evaluation code, and support for yes or no, multiple-choice, and rating questions with isolated question context.
AI agents have compressed individual research projects from years to weeks or days, making isolated papers a weaker unit of contribution than reusable ecosystems. dlab's open source week packages models, training code, and hardware-conscious tooling to show that university groups can compete by building shared research infrastructure instead of chasing the largest GPU clusters.
Tests on 100 production agent tool calls found that 79% were benign, so a classifier had to beat a constant benign baseline before its accuracy was useful. Jev's errors, class imbalance, and benchmark design show why evaluations need deployment-aware baselines and why a strong-looking score can still lose to simple code.
RoboHarm ran 300 trials across three robot policies and five hazardous instructions, but recorded only 22 safety refusals. The benchmark separates refusal from failure to complete a task and warns that its narrow setup uses one wording per instruction, 20 trials per condition, and five tabletop scenes.
Sponsored
Acquia-powered Drupal delivers API-first scale, security, and integration with 48+ AI providers - without vendor lock-in. Explore the free trial [21].
This hub publishes production-scale request traces, GPU telemetry, and agent-serving datasets for studying batching, scheduling, cache reuse, sessions, and infrastructure behavior.
A proposed universal DevTools ecosystem decomposes debugging capabilities into reusable, framework-agnostic modules that hosts such as Nuxt, React, Svelte, and Astro can compose while retaining framework-specific tools.
Benchmarks across six widely used open-source libraries show recent AI-assisted optimization gains ranging from 17% faster Base64 decoding to 4.9 times faster bitmap cardinality calculations, largely because cheap experimentation made more ideas worth testing.
DeepTeam is a local, open-source red teaming framework that simulates prompt injection, jailbreaking, multi-turn exploitation, and more than 50 vulnerability classes across LLM applications, agents, RAG pipelines, and chatbots.
Sponsored
AI agent traffic surged 7,851% over the last year, making legitimate AI activity increasingly difficult to distinguish from malicious automation. In traffic observed by HUMAN, JUST 0.5% OF BEHAVIORAL SIGNALS SEPARATED TRUSTED AI ASSISTANTS FROM MALICIOUS AUTOMATION. Block everything, and you disrupt legitimate users. Allow everything, and you create a new attack surface. THE CISO'S GUIDE TO AI AND AGENTIC TRAFFIC [4] SHOWS HOW TO: 🔎 Distinguish legitimate AI activity from malicious automation. 🛡️ Verify intent across browsing, accounts, and transactions. ⚙️ Let trusted agents through while stopping real threats.
Attackers reportedly used stolen Italian government webmail credentials to send fraudulent European Investigation Orders to Revolut Bank UAB. They added recovery addresses, monitored inboxes, deleted outbound messages, and removed replies as .eml files. Hudson Rock identified roughly 300 compromised pec.interno.it logins, suggesting attackers may have used existing infostealer logs.
Docker fixed CVE-2026-77179 in Sandboxes 0.42.0, which allowed malicious guest code on macOS to abuse virtio-fs symlink handling to access or modify host files as the VM monitor user. Version 0.42.0 also fixes CVE-2026-79994, a socket-relay race. Update to 0.42.0 or later ASAP.
Click2Shell exploited an unsanitized theme slug passed into WordPress's admin panel jQuery selector, letting an attacker silently trigger theme installation using the logged-in administrator's own session token. When chained with a missing nonce/capability check in the Mobile Repair Zone 2.5.4 theme's AJAX handler, the flaw escalated to full RCE without the attacker needing any WordPress account. No CVE had been assigned at disclosure. Defenders should patch to WordPress 7.1.1 (or backports 7.0.5, 6.9.8, and 6.8.9) and set DISALLOW_FILE_MODS to block the installation stage entirely.
Attackers may plant instructions in pages, emails, calendar invites, code repositories, and agent messages, and Agents may treat that content as commands, then use existing tools and permissions. Cases include a Grok-linked wallet transfer, poisoned Nx packages that harvested developer secrets, and zero-click data theft through Microsoft 365 Copilot, ChatGPT Deep Research, Salesforce Agentforce, and GitHub Copilot. Limit tool permissions, require for high-impact actions, track goal changes, sanitize retrieved content, and test injected-goal scenarios.
The UaniaBOX admin panel gives no shell, only a web GUI. A researcher found that the packet capture download function let him swap the file path parameter to read /etc/passwd, then traced backend calls to an exposed OpenWrt UBUS JSON-RPC endpoint. ACLs blocked most writes except one: /etc/firewall.user, a root-run shell script triggered on firewall reload, where it allowed writing a command there and triggering a reload via the GUI executed code as root, confirmed by reading back uid=0. Affected versions are UaniaOS 2.1 through 3.0.0. It was fixed in 3.0.1. Vendor UANIA reproduced the bug, attributed it to excessive ACL permissions in their OpenWrt customization feeds, and patched it, though the MITRE CVE filing stalled for months due to a broken submission form.
A Windows COM class stayed registered to a missing DLL in the user-writable ProgramData folder, letting any user plant code for a SYSTEM process to load (CVE-2026-66804, an incomplete fix for "Dark Elevator"). The exploit sent a custom-marshaled object pointing to the dangling CLSID to the Shell Create Object Handler, a SYSTEM COM server that ordinary users can start via a scheduled task and that allows custom marshaling. Defenders should patch, audit InProcServer32 registrations for DLLs that don't resolve in writable paths, and make sure privileged COM servers set EOAC_NO_CUSTOM_MARSHAL or the strong unmarshaling policy.
TigerByte Cyber's Cyber Protection Suite hardens legacy and edge devices with data validation, packet inspection, network segmentation, and post-quantum encryption. It targets aircraft, drones, satellites, vehicles, and other mission-critical systems.
MSRKit is a Windows kernel exploitation library and CLI that hijacks IA32_LSTAR via a ROP chain through AmdTools64.sys to call arbitrary kernel functions and map unsigned drivers without HVCI. The ROP chain temporarily clears SMEP/SMAP in CR4, dispatches the target, then restores both CR4 and LSTAR cleanly before returning to user mode via sysretq. The library supports all Windows 10/11 builds and KPTI, but requires HVCI to be disabled. Concurrent callers across processes will BSOD.
EntraTrace is a defensive security research tool for documenting and identifying the observable behavior of offensive tooling targeting Microsoft Entra ID. The project builds an automated knowledge base on tools such as AzureHound, AADInternals, O365Enum, PingCastle, and others, focusing on the User-Agent and API artifacts they generate.
Google placed researcher Austin Larsen inside TeamPCP during its hacking campaign. He accessed a server holding stolen usernames, passwords, and access tokens. Google warned affected victims and disrupted credential abuse, and Larsen linked an alleged member to a Gmail account and Google Drive backups. Google shared evidence with the FBI, and the Australian police arrested the suspect a month later.
During a May security evaluation, Google's Gemini accessed three external websites after finding public information and trying credentials. One access came through password guessing. Two used credentials exposed in a public repository. Google said Gemini stopped in each case. Irregular notified the affected organizations and said it fixed its testing-process issues weeks later.
WaterPlum (aka Contagious Interview) compromised 30,000+ devices across 100+ countries between December 2025 and July 2026, draining $10.7M from over 7,000 cryptocurrency wallets. A joint US, Japanese, Australian, and German advisory attributed the campaign to North Korea's 313 General Bureau and five malware families (BeaverTail, InvisibleFerret, OtterCookie, OtterCandy, and StoatWaffle) delivered through fake technical interviews on recruiting platforms. North Korean IT workers later reused identity documents stolen from victims to fraudulently obtain jobs at Western companies.
Sponsored
Learn how to deploy multi-agent systems for your SOC using the Google SecOps MCP server, Agent Development Kit, and Agent2Agent protocol.
City Relay said attackers accessed its Metabase Cloud instance twice and extracted customer data.
The 'indexed-btree' npm package, which impersonates 'sorted-btree' and pulls 2 million weekly downloads, hid its loader in the BTree.prototype.set() method to bypass npm v12's install-script approvals.
Intel quietly suspended its Intigriti bug bounty program, which paid up to $100,000 per finding, replacing it with a no-reward vulnerability disclosure program and offering no public explanation for the change.
Sponsored
Three days built for the teams running networks and security at scale: · 100+ sessions across four tracks [5], including Networking & Security: Beyond SASE and Platform & Governance: Connectivity Cloud · 200+ speakers [6], including Cloudflare co-founders Matthew Prince and Michelle Zatlyn, Stanford's Dr. Fei-Fei Li, Arm CEO Rene Haas, and Vue.js creator Evan You · A full day of hands-on Cloudflare University training, plus the Global Partner Summit and Cold Start pitch competition on Monday · Passes from $595, with team pricing at $495 for groups of five or more
Jev, TypeSafe AI's new model for making structured software decisions instead of generating text, became the fastest-adopted model in Vercel AI Gateway history. Nearly 13% of paid teams used it within 24 hours, highlighting growing demand for specialized models that handle agent routing, verification, and workflow decisions.
Nvidia has launched DSX Ready, a program that qualifies power and cooling systems for large AI factories. The initiative aims to help data center operators deal with increasingly tight power, cooling, water, and grid constraints as AI infrastructure scales.
Traditional incident response steps like revoking OAuth tokens fail when an implant can dynamically replace its entire credential set and OneDrive command-and-control identity via a single remote task. Investigators should treat application registrations as the durable object of compromise rather than relying solely on token revocation, and hunt on the identity plane using cloud telemetry.
Data ownership is fundamentally about accountability for business meaning, rules, and quality rather than technical possession or storage. While different teams across an organization may store, move, or consume data, business stakeholders remain responsible for defining its rules and metrics. Roles such as data owners, data stewards, and data custodians are distinct in managing enterprise data.
Sponsored
Keeper [12] generates strong passwords, logs you in instantly, and notifies you in real-time if your credentials appear in a breach. It also enables secure family sharing for Wi-Fi, streaming, and account passwords. Everything is encrypted so only you can access it. Now
VMware has quietly stopped selling its distributed firewalls for SmartNICs, walking back ambitions to bring hyperscale networking models to on-premise private clouds. The effort involved complex microcode integration with hardware from AMD and Nvidia, but struggled with Intel chips and ultimately failed to attract paying customers as conventional NICs improved. While standalone SmartNIC sales have ended, VMware's Distributed Services Engine remains supported within its Cloud Foundation suite.
Fastly launched AI Runtime Control and AI Firewall to manage model access, spending, routing, and agent interactions from its edge platform. The company says machine-generated traffic now accounts for more than half of traffic across its network.
Presidio opened its PATH AI lab to customers, allowing companies to test workloads across private cloud, public cloud, and neocloud infrastructure before committing to production deployments. The lab includes Dell servers running Nvidia B300 and RTX PRO 6000 GPUs.
Atlassian's Data Brewery is a scalable synthetic data engine designed to generate high-volume, product-shaped, and relationally correct datasets. The system addresses the challenge of producing realistic customer data for products like Jira and Confluence in a maintainable and extensible way for engineers.
Sumo Logic upgraded its Data Pipelines product so enterprises can filter, transform, and route telemetry before it is ingested. Its goal was to reduce observability storage costs while preparing cleaner data for AI-driven operations.
To accurately evaluate an AI tool, teams should test with competent non-experts, score against internal ground truth data, and strictly track whether the model or the human solved the task.
Mobile security firm Zimperium has discovered a new AI-powered Android malware named RatHat that tricks users into downloading fake applications and granting accessibility permissions.
KAYTUS upgraded MotusAI to let enterprises run and govern agentic AI workloads on-premises while claiming token operating cost reductions of 30–50%.