Daily brief at 7am Melbourne. Unsubscribe any time.

Wednesday 24 June 2026

Australia Is Retiring the Essential Eight — Here's What Comes Next

Australia's Essential Eight is being retired — and the replacement signals a fundamental rethink of how the ASD frames cyber defence.

Lead story

Australia Is Retiring the Essential Eight — Here's What Comes Next

The Australian Signals Directorate has confirmed it will retire the Essential Eight cybersecurity framework within the next two years. The framework — which mandates eight baseline mitigation strategies covering things like patching, multi-factor authentication, and application control — has been the backbone of Australian government and enterprise cyber posture since 2017. Retiring it is not a minor housekeeping decision. It's an acknowledgement that the threat landscape has fundamentally shifted.

The Essential Eight was designed for a world where the perimeter still meant something, and where most attacks followed predictable patterns that baseline controls could interrupt. That world is largely gone. Cloud-first infrastructure, AI-assisted intrusion, software supply chain attacks, and persistent nation-state activity have all stressed the framework past its design limits. Defenders have also complained for years that the framework's maturity model creates a checkbox culture — organisations race to hit Maturity Level 2 and declare victory, even when their actual security posture is fragile.

What replaces it? The ASD hasn't published the full replacement framework yet, but ITNews reports the successor is intended to "reflect a changing reality for security teams." That phrase is doing a lot of work. Reading between the lines, expect the replacement to lean into continuous threat-informed defence — moving away from static control lists toward dynamic risk prioritisation, detection and response capabilities, and supply chain security. The Essential Eight's current gap around detection and recovery has been a known weakness; the replacement will almost certainly address it.

The timing is significant. Australia's Cyber Security Act 2024 came into force last year, introducing mandatory ransomware payment reporting and new obligations for critical infrastructure operators. The Essential Eight was built well before any of that legislative architecture existed. Aligning the baseline framework to the new statutory environment is overdue.

Who should care most? Federal agencies and large enterprises who've built their security programmes around the Essential Eight maturity model will need to plan for transition. That's not a trivial exercise — procurement frameworks, audit schedules, vendor contracts, and board-level reporting all reference E8 maturity levels. The ASD will need to publish clear transition guidance well before the framework is formally wound down.

For smaller organisations that adopted the Essential Eight precisely because it was concrete and legible, the transition carries risk. A vague or overly complex replacement could leave a vacuum that gets filled with nothing. The ASD's job now is to make the successor just as digestible as what it's replacing — but harder to game.

What to watch: The ASD's release of the draft replacement framework, whether any maturity rating mechanism survives in some form, and how the ACSC advisory notices are updated in the interim. If you're currently mid-way through an E8 uplift programme, it's worth a conversation with your security team about whether that investment still makes sense to complete — or whether resources should pivot toward what's coming next.

Also today

Scattered Spider Members Plead Guilty Over TfL Attack

Two members of the Scattered Spider cybercrime group — aged 18 and 20 — pleaded guilty on the first day of what had been expected to be a six-week trial in the UK, admitting to infiltrating Transport for London's network in August 2024. The attack disrupted public transport services for months, exposed customer data including Oyster card refund details, and forced staff off internal systems. The guilty pleas are a rare enforcement win against a group that has previously seemed almost untouchable, having hit MGM Resorts, Caesars Entertainment, and dozens of other major organisations through social engineering. The case reinforces that even young, geographically distributed cybercrime crews can face serious legal consequences.

Krebs on Security

FFmpeg 'PixelSmash' Flaw Enables RCE via Crafted Media Files

A freshly patched vulnerability in FFmpeg's libavcodec library — dubbed PixelSmash — allows remote code execution on any application that processes untrusted media files using the library. That list is long: Jellyfin, Kodi, Emby, Nextcloud, PhotoPrism, and OBS Studio are all affected under certain conditions, with denial-of-service possible even where RCE isn't. FFmpeg is one of the most widely embedded media libraries on the planet, which makes the blast radius here unusually wide. The fix is available now; the priority for patching should be any internet-facing media server or self-hosted application that ingests user-supplied files. Australian Nextcloud and Jellyfin deployments — common in both government and SME environments — should treat this as urgent.

SecurityWeek

Fake AI Agent Skill Fooled Every Security Scanner — and Reached 26,000 Agents

Security firm AIR built a deliberately malicious AI agent skill, listed it on a popular skill marketplace, promoted it via an Instagram ad, and watched it propagate to roughly 26,000 agents — including corporate accounts. Every automated security scanner they tested marked it clean. The payload was intentionally benign (it only harvested email addresses), but the point is obvious: the AI agent skill ecosystem currently has no reliable security gate. Anyone with a marketplace listing and a small ad budget can reach tens of thousands of enterprise agents. As organisations accelerate AI agent adoption, the skill and plugin layer is emerging as the new software supply chain blind spot — one that defenders are not yet systematically monitoring.

The Hacker News

Anthropic's Fable 5 'Safe' Model Jailbroken Within Days

Anthropic's Fable 5 — marketed as a hardened, guardrailed version of its Mythos Preview model specifically designed to block use in cyberattacks — was successfully jailbroken within days of release. The restrictions preventing it from generating attack-relevant content were bypassed, though the specifics of the technique haven't been fully disclosed. It's a familiar pattern: a lab ships a model with safety restrictions, claims those restrictions are robust, and researchers or adversaries promptly prove otherwise. For defenders, the lesson is that 'safe model' labels shouldn't be taken at face value, and that any AI tool with access to sensitive systems or code should be treated as a potential attack surface regardless of its marketing.

Schneier on Security

Trump Executive Order Sets 2030 Hard Deadline for Post-Quantum Cryptography

President Trump has signed an executive order requiring US federal agencies to migrate high-value assets and high-impact systems to post-quantum cryptography (PQC) by 31 December 2030, with digital signature systems following a year later. The order covers civilian agencies; national security systems remain on a separate, classified track. The deadline matters because the 'harvest now, decrypt later' threat is already active — adversaries are collecting encrypted government traffic today, betting that quantum computers will eventually make it readable. Australia has no equivalent legislative deadline, though the ASD has published PQC migration guidance. Organisations with long data-sensitivity windows — health records, legal files, critical infrastructure configs — should be assessing their cryptographic inventory now regardless of jurisdiction.

The Hacker News

Xolis Healthtech Breach Hits 1.4 Million People via Phishing

Healthcare technology company Xsolis has disclosed a data breach affecting nearly 1.4 million individuals after attackers gained network access through a phishing email. Xsolis provides AI-driven clinical documentation tools to US hospitals, meaning the exposed data likely includes sensitive patient health information. Healthcare remains the most persistently targeted sector for data theft, both for the value of medical records on criminal markets and for the potential to extort organisations dependent on operational continuity. Australian health sector organisations should note that a breach of this nature would trigger mandatory notification obligations under both the Privacy Act and, where critical infrastructure is involved, the SOCI Act.

Bleeping Computer

Mythos AI Unearths 'Squidbleed' — a Memory Leak Hiding Since the Clinton Era

Anthropic's Mythos AI model has surfaced a memory leak vulnerability — nicknamed Squidbleed — that researchers say has gone undetected since the late 1990s. The flaw affects legacy components still in widespread use, including code paths touching NetWare, FTP, and HTTP stacks. The discovery adds to a growing body of evidence that large AI models are proving genuinely useful for finding old, subtle bugs that have slipped past human auditors for decades. It also raises an uncomfortable question: if AI is finding these bugs now, how many similar ancient flaws are still lurking in production infrastructure around the world — and who else might find them before a patch is ready?

The Register

'Cordyceps' Attack Poisons CI/CD Pipelines at Google, Microsoft, Cloudflare and More

A newly documented attack pattern dubbed Cordyceps exploits a weakness in how CI/CD workflows handle pull requests, allowing malicious contributions to execute arbitrary code with full pipeline privileges. Affected projects include Microsoft's Azure Sentinel, Google's AI Agent Development Kit, Apache's Doris database, Cloudflare's Workers SDK, and Python's Black formatter — a who's who of widely-used developer infrastructure. GitHub has already pushed an update to its actions/checkout action to block the most common exploitation paths. The real concern is that Cordyceps-style attacks are hard to spot without tooling specifically looking for them, and most organisations have neither the visibility nor the automation to detect a compromised pull request before it runs.

Dark Reading

Australia's Federal Government Establishes AI Review Committee

The Australian federal government has introduced an AI Review Committee that will provide non-binding advice to agencies on AI use and provide external scrutiny of 'AI incidents' across the public sector. The committee sits alongside Australia's existing AI Ethics Framework and the mandatory AI transparency reporting introduced under recent APS guidelines. Non-binding advice limits its immediate teeth, but the incident scrutiny function is significant — it creates a formal channel for surfacing AI failures in government systems that would otherwise be managed quietly. Whether the committee's findings are made public will determine how useful it actually is. Comparable bodies in the EU under the AI Act carry enforcement powers; Australia's version is softer, for now.

The Mandarin

Oracle Cuts 21,000 Jobs to Fund Debt-Driven AI Infrastructure Push

Oracle has laid off approximately 21,000 employees — around 10% of its global workforce — as it redirects capital toward AI data centre build-out. The company is spending billions on new compute infrastructure to position itself as a hyperscaler alternative for AI workloads, competing with AWS, Azure, and Google Cloud. The layoffs follow Oracle's aggressive debt financing strategy: it has been taking on significant leverage to fund its infrastructure ambitions. For enterprise customers already running Oracle databases and ERP systems, the question is whether reduced headcount affects support quality and product velocity — particularly for legacy systems that still form the operational backbone of many large Australian organisations.

Ars Technica

WhatsApp VBScript Campaign Installs RMM Tools — Australia Among Targets

An active malware campaign is sending fake business documents via WhatsApp Desktop and WhatsApp Web, tricking users into running VBScript files that silently install legitimate remote monitoring and management software. Once installed, the RMM tool gives attackers persistent, legitimate-looking remote access — the kind that blends into corporate environments and often evades endpoint detection. Kaspersky has confirmed Australia is among the targeted countries, alongside Malaysia, Brazil, India, Mexico, Singapore, the UK, Spain, and Taiwan. The use of legitimate RMM software as the payload is a deliberate evasion technique; defenders should review approved RMM tooling policies and alert on unexpected installations of remote access software.

The Hacker News

Sources consulted