» curated · synthesized
Skip the noise.
Read the signal.
Curated tech news and synthesis for developers and technology professionals.
» Latest posts
2695 postsRK3576 NPU on Mainline Linux: Single-Task Works, Multi-Task Stalls
Mainline RK3576 NPU driver bring-up fixes an int8 requantization bug but hits a hard wall on multi-task jobs; full root-cause ledger published.
agentproto 0.5.0 adds credential brokering, sandboxes, honest cost tracking
agentproto 0.5.0 brings credential brokering, sandboxed agent execution, and four-tier cost-source tagging to its open-source orchestration layer.
Absence Isn't Safety: Five Fail-Open Bugs in an AI Agent's Guardrails
Five fail-open bugs in an AI agent's policy engine show how missing data was misread as permission—and how each was fixed with default-deny logic.
CommitBrief — AI code reviews, right in your terminal
A provider-agnostic, local-first CLI that reviews your staged changes, a historic range, or a whole GitHub pull request. Zero telemetry, no server. Free and open source.
commitbrief.comI Red-Teamed My Own LLM Security Gateway: Every Gap, Four Passes
An engineer red-teamed his own LLM security proxy across four passes, exposing secret-leak and prompt-injection gaps — including one still open in streaming.
Microsoft's AI Bug Hunt Tripled CVEs: The Data Behind It
Microsoft's July 2026 patch hit 1150 CVEs, 3x the prior baseline. Public CVRF data reveals how AI-driven bug hunting also reshaped severity.
The hidden lock inside ArrayPool<byte>.Shared that cost 20% CPU
How live CPU profiling and BCL decompilation exposed a hidden monitor lock inside ArrayPool<byte>.Shared causing a 20% CPU regression.
When an LLM's Training Data Expires: Fixing a Car Pricing Engine
An AI car-pricing engine broke because its LLM recalled stale exchange rates, revealing risks of mandatory retrieval and cache provenance in LLM apps.
Context Bombs: Using AI Safety Guardrails to Halt Rogue Agents
Tracebit research shows context bombs hidden in canaries can trigger AI safety guardrails, cutting autonomous attacker success rates by roughly 90%.
ZVec's optimize() Can Silently Erase Crash-Recovered Vector Data
Testing shows ZVec's optimize() can silently discard crash-recovered vector data with no warning, exposing a hidden risk in embedded vector stores.
Why Coding Agents Improve More From Harness Design Than Models
LangChain pushed a coding agent to Terminal-Bench's top 5 without changing the model — the gains came entirely from harness design.