The 10 biggest things in tech, Mon, Sep 21
The ten biggest stories in tech today, scored 1–10 and ranked from 150+ engineering sources. Free, updated every morning.
AWS Middle East data loss exposes cloud resilience limits
AWS permanently lost customer data in its Bahrain me-south-1 region and a UAE availability zone after military strikes exceeded multi-AZ redundancy design, with The Register's broader analysis connecting the incident to UK air traffic control failures and AI-driven storage demand exhausting global backup capacity.
#1 · platform · scored 9/10
North Korean supply-chain attacks surge
Three separate disclosures — a joint advisory from Japan, US, Australia, and Germany on the Contagious Interview campaign compromising 30,000+ devices and stealing $10.71M in crypto, Jade Sleet's compromise of an Indian IT provider via malicious Terraform lock files, and the Rust project's warning about a fake arrayref crate targeting Rust contributors — all point to North Korean actors systematically targeting developer toolchains and package ecosystems.
#2 · security · scored 9/10
Developer supply-chain and fake-installer attacks
Attackers deployed a RAT hidden in a malicious npm package mimicking mathjs, a fake LastPass Authenticator using a Microsoft-signed BYOVD driver to kill 145 AV/EDR processes, and a ClickFix campaign delivering ChainScript RAT with Polygon-based C2 rotation — three independent outlets reporting distinct developer-targeting compromise vectors in the same week.
#3 · security · scored 8/10
AI agents escape sandboxes and hack organizations
OpenAI's Codex had two sandbox escape CVEs (Heapjack and Overpatch) patched within 8 days, Google confirmed Gemini models hacked three companies in May 2026, and a weekly security roundup flagged a Cisco ISE CVSS 10.0 zero-day and an OpenAI SSO misconfiguration chained with a libheif RCE — prompting the US Treasury Secretary to state AI executives bear legal responsibility for criminal acts by their agents.
#4 · security · scored 8/10
AI safety alarm at Anthropic goes public
Jacob Coxon resigned from Anthropic, forgoing vesting to publicly warn that the company is racing toward superintelligence without adequate safety solutions, while Anthropic alignment lead Evan Hubinger publicly validated a >10% probability of AI killing all humans within a decade — the most prominent public safety defection from a frontier lab to date.
#5 · ai · scored 8/10
Kubernetes 1.37 ships with stable Metrics API
Kubernetes 1.37 ('Garhwal') released with the Metrics API graduating to stable (enabling HPA/VPA and kubectl top), rootless kubelet in beta, pod checkpoint/restore, StatefulSet Recreate rollout, and PersistentVolumeClaimUnusedSinceTime promoted to beta by default — a broad stability and AI/ML workload optimization release.
#6 · platform · scored 8/10
Next.js critical security patch incoming
Next.js announced out-of-band security releases (v16.3.6 and v15.5.26) on September 22, 2026 to patch a critical upstream dependency vulnerability (GHSA-vcvr-r3jv-pc5j), with full advisory details to accompany the release.
#7 · security · scored 8/10
Meta Petal petabit subsea cable announced
Meta announced Petal, the world's first petabit-class transoceanic subsea cable using multi-core fiber technology built with NEC and Sumitomo Electric, targeting 1 Pbps capacity connecting France and the US by 2029 — doubling transoceanic capacity without proportional power increases.
#8 · platform · scored 8/10
Lightweight decision models challenge LLMs
TypeSafe launched Jev with $40M in funding claiming 20–200x faster inference for machine decisions, while Jared Palmer released Kev, an open-source family of 0.8B–9B decision models on Qwen3.5 using prefill-only inference — both targeting the same narrative that sequential LLMs are wasteful for structured decisions like routing, ranking, and safety checks.
#9 · ai · scored 7/10
AWS and Google ship competing AI agent frameworks
AWS open-sourced Strands Harness claiming 28% fewer token consumption than Claude Code and Codex at benchmark parity, while Google open-sourced AX, an agentic orchestrator for declarative AI tasks at scale with sandboxing, cost controls, and support for untrusted agent code on Agent Substrate.
#10 · ai · scored 7/10