Daily brief at 7am Melbourne. Unsubscribe any time.

Sunday 12 July 2026

A Poisoned npm Package Dropped a Rust Infostealer Before Anyone Had Time to Blink

A poisoned npm package dropped a cross-platform infostealer on install — and the supply chain security story is even wilder than the attack itself.

Lead story

A Poisoned npm Package Dropped a Rust Infostealer Before Anyone Had Time to Blink

Someone compromised the jscrambler npm package and pushed version 8.14.0 with one simple goal: get malicious code running on your machine the moment you type npm install. The attack used a preinstall hook — a legitimate npm feature that runs a script before the package even finishes landing on your disk — to drop and execute a compiled Rust binary. Three flavours: Windows, macOS, Linux. Whoever wrote it wasn't in a hurry to exclude anyone.

What makes this one sting is that jscrambler is not some obscure utility you'd only find in a hobby project. It's a JavaScript obfuscation and protection tool used by developers who are actively trying to make their code harder to tamper with. The irony of an infostealer hiding inside a security tool is the kind of thing that writes itself.

Socket, the supply chain security firm, flagged the malicious release just six minutes after it was published. That's genuinely impressive detection speed — but six minutes is still enough time for automated CI/CD pipelines to pull the package, install it, and have credentials or session tokens silently exfiltrated before a human being is in the loop. Modern software pipelines don't wait for approval. They just run.

Rust is increasingly the language of choice for malware authors targeting multiple platforms from a single codebase. It compiles to small, fast, hard-to-analyse binaries, and it doesn't carry the Python or Node runtime signatures that some endpoint tools are tuned to catch. If you're writing a cross-platform infostealer in 2026, Rust is a sensible call.

The broader supply chain picture is uncomfortable. npm's publish model has very low friction by design — that's what made the JavaScript ecosystem explode. But low friction in both directions means attackers who get hold of a maintainer's credentials (or compromise a CI token) can push a poisoned release in seconds. The jscrambler team presumably still holds the package name, the version history, the reputation. All of that equity becomes the attack surface.

For Australian developers and organisations, the exposure here is the same as anywhere else: any project with jscrambler in its dependency tree that ran an install or update in the window between publication and Socket's flag should be treated as potentially compromised. The Australian Signals Directorate's Essential Eight specifically calls out patch and software management — but "only install from trusted sources" is harder guidance to action when the trusted source itself gets flipped.

What to do now: check your lockfiles. If jscrambler@8.14.0 appears anywhere in a package-lock.json or yarn.lock, treat the environment as compromised and rotate credentials. Audit your preinstall hooks more broadly — they're a persistent blind spot in most dependency review processes.

The jscrambler team has since pulled the malicious version, but the underlying problem — that npm publish is fast, easy, and only lightly gated — isn't going anywhere. Supply chain attacks are not getting less common. They're getting more precise.

Also today

CISA Wrote Its Incident Playbook During the Actual Incident

The US Cybersecurity and Infrastructure Security Agency has revealed it had no formal internal incident response playbook when a contractor employee uploaded reams of passwords to a publicly accessible GitHub repository earlier this year. CISA built the playbook as it responded to the breach — hardly an advertisement for the agency charged with telling everyone else how to handle incidents. The disclosure came via independent journalist Brian Krebs, who originally broke the story in May after a GitGuardian researcher spotted the exposed credentials. It's a reminder that even the organisations writing the security guidance aren't always following it — and that Australia's ASD faces similar institutional pressures to practise what it preaches.

TechCrunch

Ghost Accounts Are Quietly Mapping GitHub Organisations at Scale

Multiple campaigns are using dormant or throwaway 'ghost' accounts to systematically enumerate GitHub organisations — harvesting repository names, member lists, and structural metadata via the GitHub API. The reconnaissance is low-noise by design: ghost accounts generate minimal activity signals, and the GitHub API is generous with what it returns to authenticated users. The mapped intelligence likely feeds downstream spearphishing or targeted repository attacks. For any organisation that keeps internal tooling, infrastructure-as-code, or product code in GitHub — which in 2026 is most of them — this is a reminder that your org's public-facing GitHub footprint is itself an attack surface worth auditing.

SecurityWeek

Critical Zimbra XSS Flaw Turns a Crafted Email Into a Code Execution Vector

Zimbra is urging customers to patch immediately after disclosing a critical stored cross-site scripting vulnerability in its Classic Web Client. An attacker can craft a malicious email that, when opened, executes arbitrary JavaScript inside the recipient's authenticated session — meaning full account access without needing credentials. No CVE has been assigned yet, which means automated scanning tools won't catch it by identifier alone. Zimbra remains widely deployed in government, education, and enterprise environments across the Asia-Pacific region, and organisations running self-hosted instances are particularly exposed because patching timelines are entirely in their own hands.

The Hacker News

China and India-Linked Actors Both Targeted Pakistani Police in Overlapping Espionage Campaigns

Researchers have detailed two years of sustained cyber espionage against Pakistani law enforcement, including the Balochistan Police, carried out by threat actors aligned with both China and India. The compromised assets included web application servers managing criminal records and citizen data — exactly the kind of intelligence useful for tracking activists, dissidents, or persons of interest across borders. What makes this disclosure notable is the overlap: two separate nation-state actors, competing geopolitical interests, hitting the same target simultaneously. It's a sharp illustration of how a single under-resourced government network can become a contested intelligence battleground.

The Hacker News

AI Finds a 15-Year-Old Root Privilege Bug in Linux That Human Reviewers Missed

An AI-assisted code analysis tool has uncovered a 15-year-old privilege escalation vulnerability in the Linux kernel — one that human code reviewers walked past for a decade and a half. The flaw allows a local attacker to gain root access, which in shared or multi-tenant environments is a serious problem. The finding adds to a growing body of evidence that AI-assisted vulnerability research is genuinely surfacing bugs that traditional auditing misses, not just rediscovering known issues. It also raises an uncomfortable question: if AI can find these now, how long before adversaries are routinely using the same tools to find them first?

WIRED Security

'Squidbleed': A 29-Year-Old Squid Proxy Bug Can Leak HTTP Requests

Bruce Schneier has flagged a newly disclosed vulnerability in Squid, the widely deployed open-source proxy server, that has apparently been sitting in the codebase for 29 years. The flaw can leak HTTP request contents — a significant concern in environments where Squid proxies internal traffic and the leaked requests might contain session tokens, credentials, or sensitive query parameters. Squid is common in enterprise and government network architectures, including in Australia where it appears in university, ISP, and agency deployments. A nearly three-decade-old bug surviving this long in a security-adjacent component is the kind of finding that makes you wonder what else is hiding in mature C codebases.

Schneier on Security

Quantum Error Correction Can Now Continuously Recalibrate the Processor It's Protecting

New research published this week demonstrates that quantum error correction can be used not just to catch computation errors, but to continuously recalibrate the underlying quantum processor in real time — using the error information itself as a feedback signal for reinforcement learning. It's a meaningful step toward practical quantum computing: one of the field's stubborn problems has been that quantum hardware drifts out of calibration faster than useful computation can occur. Using the error correction layer to simultaneously fix that drift is an elegant solution. It doesn't put a general-purpose quantum computer on anyone's desk, but it removes one of the more fundamental engineering obstacles.

Ars Technica

Meta Pulls Instagram AI Feature After Creators Push Back Hard

Meta has quietly removed an AI feature from Instagram after a swift and vocal backlash from creators. The feature — which allowed users' public content to be referenced by an AI tool — was framed as a creative aid, but many creators felt it amounted to using their work to power a product without meaningful consent. Meta's response was swift by its standards: the feature was pulled within days of the criticism peaking. The episode is another data point in a pattern where AI products are shipped, encounter creator or user backlash around consent and data use, and get walked back — a cycle that seems likely to repeat as platforms integrate generative AI more deeply into their products.

TechCrunch AI

OpenAI Wants ChatGPT in Every Room of the House

OpenAI is hiring a dedicated product manager to build ChatGPT experiences specifically for families, caregivers, and older adults — a clear signal that the company is moving beyond productivity tools and into the ambient, domestic layer of everyday life. A household-focused ChatGPT product would compete in territory currently occupied by Amazon's Alexa and Google's Home ecosystem, but with a conversational depth those platforms have never matched. The play makes strategic sense: household adoption locks in habitual usage and creates natural upsell pathways to paid tiers. It also raises fresh questions about AI in contexts involving children and vulnerable adults — areas where Australia's Online Safety Act and the Privacy Act's health and sensitive information provisions would be directly relevant.

TechCrunch AI

Bluesky Has a Permanent CEO Now

Toni Schneider, who has been running Bluesky in an interim capacity, has been confirmed as the platform's permanent CEO. Schneider brings a credible operational background — he previously ran Automattic, the company behind WordPress.com — and has said he is fully committed to the decentralised social platform's unconventional model. Bluesky has grown substantially since Twitter's implosion accelerated in 2023, attracting users who wanted a chronological, algorithm-optional feed. Confirming a permanent chief is a maturity signal for a platform that has spent much of its existence being described as a 'Twitter alternative' rather than a product in its own right.

TechCrunch

Phia, the Bill Gates-Adjacent Shopping App, Accused of Cookie Stuffing

Phia, the shopping startup co-founded by Phoebe Gates (daughter of Bill) and Sophia Kianni, is facing allegations of 'cookie stuffing' — a form of affiliate fraud where a website plants tracking cookies for purchases it had no role in generating, then collects the resulting commission. A Bloomberg investigation found the practice allowed Phia to claim affiliate credit for transactions its app didn't actually influence. Cookie stuffing is an old trick in affiliate marketing, but its appearance in a well-funded, high-profile consumer startup is notable. The FTC has taken enforcement action against cookie stuffing before; whether this rises to that level remains to be seen.

TechCrunch

Sources consulted