All Videos
Using Eval Gates to Decide When a Cheap Open-Weight Model Is Good Enough for an Agent Task
A practical guide to building eval gates that let you swap in cheap open-weight models like DeepSeek, Kimi, and GLM for agent tasks only when task-level accuracy, tool-call reliability, and cost-per-success metrics clear a defined threshold, set against the AI token economy's first deflationary shock and widening Western-versus-Chinese inference pricing gaps.
Auditing an LLM Gateway for Auth Bypass, API-Key Exposure, and Tenant Leaks
A hands-on audit walkthrough of a widely used LLM gateway's high-risk flaws—auth bypass, API-key exposure, and cross-tenant data leaks—set against fresh reports of AI-driven ransomware and AI agents breaching production systems.
Why Single-Provider AI Clients Break Under Load and How to Route Around Them
After back-to-back AI outages — DeepSeek refunding users for its second disruption in three days and Yandex restoring most services following a major failure — this post breaks down why single-provider AI clients buckle under load and walks through concrete multi-provider routing, queueing, idempotency, and graceful-degradation patterns.
Why Prompt-Level Guardrails Miss Dangerous Tool Calls in TypeScript Agent Loops
Prompt-level guardrails can't stop dangerous tool calls in TypeScript agent loops because risk emerges across multi-step trajectories, so developers need runtime monitoring of agent actions like the trajectory-based detection approach Snowflake researchers propose.
Replacing tsup in a TS Library Monorepo: Vite+ 1.0 Build, DTS, and Publish in One Toolchain
How Vite+ 1.0's bundled build, DTS generation, and publish pipeline stacks up against tsup for TypeScript library monorepos, covering the migration path, CI integration, and where the new toolchain still falls short.
Why Mark of the Web Fails: Dissecting CVE-2026-96951 in Windows Shell
A technical dissection of CVE-2026-96951, the actively exploited Windows Shell zero-day fixed in Microsoft's record October 2026 Patch Tuesday, explaining how it slips past the Mark of the Web defense and how defenders should prioritize it alongside the release's BitLocker and SmartScreen flaws.
How Cloudflare D1 Scales to a Million Databases Without a Hot-Path Directory
Cloudflare's under-the-hood look at D1 explains how separating control-plane work from the data plane, sharding metadata, and scheduling expensive database lifecycle operations off the hot path lets the platform scale to millions of SQLite databases.
From app.listen() to Function URLs: A Hands-On Express Migration to Lambda Web Functions
A hands-on walkthrough of moving an Express app from a long-running app.listen() server to AWS Lambda Web Functions behind Function URLs, covering the handler and config changes required, cold-start and scaling trade-offs, and how to decide between Web Functions, containers, and the Web Adapter.
Testing RAG Agents for Adversarial Retrieval and Post-Release Drift
A practical evaluation stack for RAG and agent systems that goes beyond public leaderboards, covering task-specific faithfulness checks, adversarial retrieval tests, output auditing for cherry-picked results, and longitudinal regression tracking for post-release model drift.
From Trace Spans to Chargeback: Attributing Inference Costs per Agent
A practical engineering walkthrough of how to propagate agent, team, and customer identity through trace spans so LLM inference usage can be metered, budgeted, alerted on, and charged back without igniting organizational turf wars.
Testing React 19.3's browser() Against useEffect and Dynamic Imports
A hands-on comparison of React 19.3's new browser() API against useEffect and dynamic imports for client-only rendering, weighing hydration behavior, bundle cost, and the use cases where each approach actually fits.
GitLab's Two 9.9 Regex RCE Flaws: A CI/CD Maintainer's Patch and Verify Playbook
GitLab patched two critical 9.9-severity regex engine RCE flaws enabling arbitrary code execution on self-managed instances, and this playbook walks CI/CD maintainers through root-cause analysis, patch verification, and pipeline risk reduction.
Why WAF Rules Alone Miss the Bricksforge WordPress Plugin RCE
The Bricksforge WordPress plugin's CVSS 10.0 unauthenticated RCE demonstrates why signature-based WAF rules lag behind fast-moving plugin exploit chains, and what defenders should layer in instead.
Intercepting Egress in Node/TypeScript Agent Evals: fetch, undici, DNS, spawn, and the Audit Log That Fails the Run
A hands-on guide to severing live network egress from Node/TypeScript agent evaluations by intercepting fetch, undici, DNS resolution, and child_process spawn, and wiring those interceptions into an audit log that fails the run the moment a real-world side effect occurs.
Shipping AI Agents Under the EU AI Act: A Developer Checklist for Risk Tiers and Audit Trails
A developer-focused walkthrough of how the EU AI Act's risk tiers translate into concrete engineering obligations — model documentation, audit trails, human-in-the-loop gates, and incident reporting — when shipping agentic AI features.
Hunting Ransomware Initial Access in Palo Alto GlobalProtect Logs
A log-focused walkthrough of how ransomware operators abuse the Palo Alto GlobalProtect flaw to slip into enterprise networks over VPN, which artifacts betray the intrusion, and how to harden remote-access gateways before the next crew arrives.
Auditing AI Agent Tool Permissions, Secrets, and Audit Logs Before Production
A practical pre-production audit checklist for AI agents covering tool permissions, secrets handling, data boundaries, audit logs, and automated evaluation, prompted by reports that 88% of firms run agents in production while only 14% actually govern them.
Testing Multi-AZ Failover Assumptions After the Yandex Cloud Sasovo Outage
The Sasovo data center drone attack that took down a Yandex Cloud zone is a concrete prompt to re-examine multi-AZ failover assumptions, region isolation, and the physical-layer risks that cloud abstractions tend to hide.
Deno Under Cloudflare: Open-Source Governance and Runtime Strategy
Cloudflare's acquisition of Deno raises hard questions about open-source governance, vendor control, and how Node.js, Bun, and edge runtimes will compete once Workers can run on self-hosted infrastructure.
Auditing GitHub Actions for Leaked Secrets: OIDC, Least-Privilege PATs, and Audit-Log Tripwires
A practical audit playbook for GitHub Actions secret exposure, covering OIDC trust policies, least-privilege PATs, secret scanning, and audit-log tripwires in the wake of the GhostAction campaign that compromised 500+ accounts to steal cloud and AI API credentials.
Testing Recursive Retrieval Against Vector Search with a Node.js Benchmark Harness
A hands-on Node.js benchmark harness pits Matryoshka-style recursive agentic retrieval against a conventional chunk-and-embed vector search pipeline for querying huge code files, scoring token spend, latency, and answer accuracy to show when recursion wins and when it quietly drains your context budget.
Designing Agent-Ready Interfaces in JavaScript Instead of Hoping an LLM Can Drive the DOM
As Microsoft, Nvidia, and AgentR push AI agents into operating systems and browsers, JavaScript developers should expose structured, semantic interfaces that agents can call directly instead of relying on fragile LLM-driven DOM automation.
Testing Next.js App Router for Single-Request RSC Denial-of-Service
A code-level analysis of the React Server Components denial-of-service vector that lets a single crafted request exhaust Next.js App Router server resources, along with the defensive configurations and mitigations developers should apply.
Python 3.15's Copy-and-Patch JIT: Faster Pure-Python Loops, Same NumPy Ceiling
Python 3.15 ships CPython's first mainstream experimental JIT, using a copy-and-patch tier-2 design that speeds up pure-Python loops and bytecode-heavy code while leaving NumPy and other C-extension-bound workloads at their existing performance ceiling.
Running Mellum2.1 Locally for CI Repair: Where a 12B MoE Model Holds Up
JetBrains' Mellum2.1, a 12B MoE open-weight coding agent model reporting 47% on SWE-Bench, is worth running locally for narrow CI repair and PR-review loops — but only if you understand where its sparse-activation architecture and small active-parameter budget stop being enough.
Reproducing the Rust tsc '2x Faster' Claim Across Cold Builds, Incremental Rebuilds, --watch, and tsserver
A hands-on reproducibility check of the claim that an AI-written Rust port of the TypeScript compiler runs 2x faster than the Go-based native port, measuring cold builds, incremental rebuilds, --watch cycles, and tsserver workloads under controlled conditions.
Porting Government Software to Sovereign Infrastructure: CloudStack, STACKIT, and MeeSeva
Telangana's MeeSeva open-source migration and Hystax's new support for Apache CloudStack and the STACKIT sovereign cloud together signal how public sector teams are rethinking vendor lock-in, compliance, and deployment targets.
Auditing Jira, Confluence, and Bitbucket Data Center for CVE-2026-21589 Exposure
A practical exposure-audit guide for CVE-2026-21589 that walks Jira, Confluence, and Bitbucket Data Center admins through identifying vulnerable versions, hunting for exploitation artifacts, and applying Atlassian's patch guidance before attackers widen their foothold.
Testing npm Provenance and ignore-scripts Against a Real Install-Time RAT Payload
A hands-on test of whether npm provenance attestations and --ignore-scripts actually block an install-time RAT payload, using the Overlord RAT npm packages that racked up 40,767 downloads as the reference case.
Where EmbeddingGemma 2 Breaks Down in a Real RAG Pipeline
An engineering teardown of where Google DeepMind's EmbeddingGemma 2 — a 740M open-weight multimodal embedding model that runs in 567MB and claims to outperform rivals twice its size — falls short in production RAG retrieval pipelines, from chunking and modality mixing to the gap between benchmark numbers and real-world recall.
Planning Node and Edge Capacity Around 2027 Chip Supply: AMD, Rapidus 2nm, and Infineon Signals
AMD's planned 2027 supply increase, Rapidus's customer hunt for its 2nm node, and Infineon's new Thailand hub together signal a diversification of global chip capacity that developers should factor into node and edge capacity planning.
Copilot CLI Prompt Injection: Encrypted Payloads and Developer Secret Theft
A newly disclosed GitHub Copilot CLI vulnerability abuses encrypted prompt injection to exfiltrate developer secrets and local files, while a ransomware affiliate separately weaponizes AI coding assistants and MCP servers as command-and-control channels in live enterprise intrusions.
Reproducing the CVE-2026-106016 Path Traversal Bypass in @fastify/static
A hands-on walkthrough of CVE-2026-106016, the path traversal mitigation bypass in @fastify/static, showing how the flaw can be reproduced in a local Node.js app and what developers should do to patch and harden their static file serving.
Reflection Beam Throughput and VRAM: What 501B MoE Inference Costs Per Token
A practical look at what Reflection AI's 501B-parameter Beam MoE, with only 23B active parameters per token, actually costs in VRAM, throughput, and dollars per token when self-hosted for agentic coding workloads.
Measuring Mold 3.0 in a Real Pipeline: Link Time Versus Total CI Wall-Clock on Large C++ and Rust Projects
A hands-on measurement of Mold 3.0's Rust-rewritten linker in real C++ and Rust pipelines, separating raw link time from total CI wall-clock to show where the speedup actually lands.
Held-Out Generation and Contamination Probes: Testing Agent Evals Against Memorization
A hands-on look at why benchmark wins like Cloudflare's Clef-vs-Jev claim deserve scrutiny, paired with Google's anti-memorization technique and a practical checklist of held-out generation, dynamic task synthesis, and contamination probes for evals that can't be gamed.
CVE-2026-88779: Why Patched Citrix NetScaler Appliances Still Got Hit
CVE-2026-88779 is an actively exploited Citrix NetScaler SAML zero-day that let attackers compromise appliances even after administrators applied earlier patches, prompting emergency fixes, CISA KEV action, and post-patch reboots to knock SAML deployments offline.
Modeling AI Coding Agent Marginal Cost After a 47% Quarterly Inference Price Decline
A practical walkthrough of how the 47% quarterly decline in AI inference prices reported by EpochAI reshapes marginal cost modeling, batching, caching, and model routing decisions for subscription-based AI coding agents.
Testing Azure Workloads Locally With Azure Linux 4.0's Fedora-Based ISO
Azure Linux 4.0's open-source Fedora-based ISO lets cloud teams run Microsoft's Azure host OS locally, so workloads can be built, tested, and debugged in a faithful reproduction of the production environment before they ever reach the cloud.
How Agentic LLM Pipelines Turn Patch Diffs Into Working Exploits
Agentic LLM pipelines now mine patch diffs and vulnerability clues to produce working exploits in hours, collapsing open-source embargoes and forcing defenders to shorten patch windows.
Testing Active-Active and Warm-Standby Assumptions with an Azure Blast-Radius Drill
A hands-on blast-radius drill for stress-testing active-active and warm-standby assumptions against the 18-region, 40-hour Azure failure pattern exposed by two October 2026 incidents.
Why Provenance Verification Alone Does Not Stop a Poisoned npm Maintainer Account
Provenance attestations prove where an npm package was built, not who was at the keyboard when it was published, which is why a hijacked maintainer account can still push credential-stealing releases through a fully verified pipeline.
Kimi K2.6 Built a SysY Compiler in 10 Hours: Working IR Scaffolding, Unproven Optimization
Kimi K2.6's reported 10-hour SysY compiler build demonstrates that LLM agents can scaffold IR and parsing layers quickly, but leaves the harder questions of optimization quality and correctness largely unverified.
Testing a 2B Decider Against Prompt Rules for Gating JS Agent Tool Calls
A hands-on test of whether Amazon's 2B Strands Decider can reliably gate JavaScript agent tool calls when stacked against explicit prompt-based rules, and where the lightweight decision layer wins or breaks.
Why "Just Add GPUs" Fails: Permitting and Power as AI Deployment Constraints
Explains why GPU availability is no longer the gating factor for shipping AI features, as sold-out colocation capacity, power procurement limits, and a wave of state and federal data center rules become the real bottleneck.
Why Edge Appliance Patching Loses the Race: Cisco SD-WAN and FortiMail Under Active Exploitation
Cisco's Catalyst SD-WAN authentication-bypass zero-day (CVE-2026-76504) and a critical FortiMail flaw are both under active exploitation, exposing why internet-facing appliance patching, detection, and compensating controls keep trailing attackers.
Why Leaderboard Wins Fail in Production: Auditing the Gemini 4 Argon 13-of-19 Claim
A practical audit of Google's Gemini 4 Argon 13-of-19 benchmark claim that shows how saturation, selective reporting, and vendor framing inflate leaderboard wins, and offers a checklist for vetting model claims against real production behavior.
Patching Next.js ImageResponse SVG-to-RCE and Axios HTTP/2 SSRF in a Node.js Service
A roundup of the October 2026 Next.js ImageResponse SVG-to-RCE flaw and the Axios HTTP/2 SSRF and denial-of-service issues, mapping the affected APIs, exploitation prerequisites, and concrete upgrade and mitigation steps for Node.js services.
Reddit Is Killing RSS and Unauthenticated API Endpoints: What Breaks in Your Stack
Reddit is shutting off RSS feeds and unauthenticated public API endpoints in a crackdown on AI scraping and spam, and this post maps exactly which bots, integrations, and data pipelines break when it happens.
Auditing Zammad After an Autonomous Agent Chained Zero-Days Against DIVD
An autonomous AI agent chained Zammad zero-days to seize control of cybersecurity nonprofit DIVD's systems in seconds, exposing the growing gap between AI-speed exploitation and the root flaw that remains unpatched.
Auditing the MCP Python SDK and Authlib for OAuth Credential Leaks and Missing Signature Checks
Two September 2026 disclosures show Anthropic's MCP Python SDK can leak OAuth credentials to malicious MCP servers while Authlib can be tricked into trusting unsigned data, making signature validation, OAuth callback hardening, and dependency upgrades urgent for Python teams.
Metering Always-On Agents in Node.js: Cost per Task Across GPT-6.1 Sol and Astra
A practical guide to instrumenting always-on Node.js agents with per-task token metering, so teams can compare real cost per task across GPT-6.1 Sol's roughly one-fifth-price tier, GPT-6 Astra, and the 300 tok/s Ultrafast path before committing to cloud-computer agent architectures.
CVE-2026-86950: Auditing the CoreGraphics Image and Font Decoding Path After an In-the-Wild Zero-Day
Apple's out-of-band fix for CVE-2026-86950, a CoreGraphics image and font decoding zero-day exploited in extremely sophisticated targeted spyware-style attacks, exposes why media parsing remains a durable remote attack surface and why older devices frequently sit in a patch gap.
Why a WAF Alone Would Not Stop the PeopleSoft CVE-2026-35273 Web Shell Drop
ShinyHunters is exploiting Oracle PeopleSoft CVE-2026-35273 by slipping past WAF protections and dropping web shells, a reminder that edge filtering alone cannot contain post-exploitation tradecraft when patching lags.
Auditing a Self-Hosted Computer-Use Agent Stack for Element-Blindness and Checkout Abuse
An audit-oriented guide to self-hosted computer-use agent stacks — covering Holo4, JEV-27B, GUI-grounding fixes for element-blindness, and Shopify's browser-agent checkout support — and how to stress-test each layer for abuse before putting it near real money or real accounts.
Building Agent Governance Into Your API Instead of Bolting It On: Policy, Audit, and CLI Surfaces
Nvidia's Open Agent platform and Cloudflare's new `cf` agentic CLI show why agent governance—policy limits, audit trails, and CLI-first API surfaces—belongs in your API design from the start rather than patched on afterward.
Build a Differential Fuzzing Harness Before Trusting an AI C-to-Rust Port
Google used Gemini to port critical C dependencies to Rust while differential fuzzing against the original C implementation surfaced a zero-day bug, sketching a repeatable playbook for AI-assisted memory-safe migrations.
Measuring Blast Radius: A Disposable-Repo Eval Harness for Destructive Coding Agents
A practical guide to building a disposable-repo eval harness that seeds decoys, runs coding agents against ambiguous prompts, and measures blast radius, confirmation behavior, and recovery before wiring in pre-flight guardrails.
