Microsoft researchers demonstrated that editing fewer than ten neurons can reliably backdoor an AI agent, exposing why cryptographic weight hashes and provenance checks cannot detect tiny malicious weight edits in open-weight model supply chains.
A deep technical walkthrough comparing AWS Emulator, LocalStack, and Moto in one test suite to show which AWS checks survive local emulation, where service-fidelity gaps appear, and when pre-deploy validation is safe enough for CI.
A technical walkthrough of the actively exploited Outlook NTLM credential-theft zero-day, covering the exploit path, suspicious Outlook child processes, outbound SMB authentication detection, and mitigation priorities beyond patching.
Angular's TypeScript 7-powered compiler beta introduces a Rust-based preprocessor, and this audit walks through what it replaces in the build pipeline, how diagnostics and source maps hold up against the legacy compiler, and where transformer and beta-stage compatibility risks bite enterprise…
Angular's TypeScript 7-powered compiler beta is reported to move part of the build pipeline into a native preprocessor, described as Rust-based, while Microsoft's publicly documented TypeScript native port is a Go codebase. This audit skips the faster-builds headline and measures the three properties that actually decide whether a real workspace can adopt the beta: diagnostics parity, source-map fidelity, and transformer compatibility. We build a fixture workspace with one application and one publishable library, strict templates everywhere, and template checks promoted from warnings to errors so the diagnostic surface is diffable at all. Both compilers run from the same lockfile and are selected by environment variable, with exact versions pinned and no carets, so the control can't quietly evaporate between machines. Then we look at where the native-to-JavaScript seam leaks: byte offsets becoming line and column numbers, UTF-16 code units versus UTF-8 bytes, spans expressed as start-length or start-end, and spans that cross the synthetic template boundary. None of those failures crash a build or turn CI red, which is exactly why they matter. Full write-up: https://hackyjs.com/posts/auditing-angulars-typescript-7-beta-diagnostics-parity-source-map-fidelity-and-transformer-compatibility
Most agent bills come from the loop, not the model. This short runs the arithmetic from the post: a ten iteration loop sends 69,100 input tokens against 1,800 output tokens, because every iteration re-sends the system prompt, the tool schemas and every tool result so far. Full write-up: https://hackyjs.com/posts/designing-agent-tool-loops-around-per-call-cost-instead-of-frontier-model-assumptions What's covered: - the four input token buckets, and why tool schemas are a fixed per-call tax - a per-call meter that logs input, cached and output tokens plus wall time - splitting the planner from the executor, with loop caps and escalation criteria - why per-call timeouts don't compose: 9.6 percent at ten calls, 55 percent at eighty Note: the post treats Muse Ember's parameter count, context window, pricing and benchmarks as unknown. The numbers here are loop arithmetic you can rerun against any candidate model, not model benchmarks.
An exploited Outlook zero-day leaks Windows NTLM authentication hashes in the November 2026 Patch Tuesday release. This short walks the defensive audit path: what the Net-NTLMv2 proof is, why password rotation does not fix a captured response, how relay targets like ADCS change severity, and which controls break the relay. It closes with a triage order for a roughly 1,094-CVE release. Full write-up: https://hackyjs.com/posts/auditing-outlook-ntlm-hash-exposure-from-zero-day-path-to-patch-tuesday-triage