Daily brief at 7am Melbourne. Unsubscribe any time.

Thursday 18 June 2026

FortiBleed: 73,000 Fortinet Credentials Dumped — and Yours Might Be Among Them

Tens of thousands of Fortinet firewalls have had their credentials stripped bare, a Queensland sugar mill is counting the cost of a mid-harvest cyberattack, and G7 leaders are quietly panic-buying sovereignty from American AI.

Lead story

FortiBleed: 73,000 Fortinet Credentials Dumped — and Yours Might Be Among Them

A dataset dubbed "FortiBleed" surfaced this week containing what appears to be working VPN credentials for 73,932 Fortinet and FortiGate firewall URLs across organisations worldwide. The list includes names you'd recognise: Oracle, Lenovo, FedEx, and at least one NATO contractor. Researchers at SOCRadar separately confirmed roughly 30,000 actively compromised Fortinet devices. If you manage Fortinet infrastructure and haven't rotated credentials yet, that sentence should have your full attention.

The mechanics matter here. This doesn't appear to be a single new zero-day — it looks more like an aggregated harvest from multiple sources: older CVEs that were patched but whose credentials were never rotated, plus active exploitation of two critical FortiSandbox vulnerabilities (CVE-tracked, CVSS scores in the 9+ range) that Fortinet patched back in April. Multiple security firms have now observed in-the-wild exploitation coming from several independent threat actors, which means it's not a coordinated campaign — it's a free-for-all.

The credential dump is the nastier part. Even organisations that patched quickly may have left working credentials in the wild. Credentials exfiltrated before a patch don't expire with the patch. Anyone holding a valid username/password can still walk through the front door.

Fortinet gear is deeply embedded in Australian enterprise and government networks. The Australian Signals Directorate has previously flagged Fortinet vulnerabilities in its high-priority advisory cycles, and Fortinet sits on the list of approved network appliances across multiple federal and state government procurement frameworks. SOCI Act entities in particular — critical infrastructure operators in energy, water, transport, and comms — should treat this as a first-priority rotation event, not a scheduled maintenance item.

What should you do right now? Rotate all Fortinet VPN credentials immediately, whether or not you believe you were patched. Check your FortiSandbox, FortiGate, and associated device logs for any authentication events from unusual geolocations or off-hours timestamps going back to April. If you're using the same credential sets across multiple devices — a very common operational shortcut — treat this as a full network-wide rotation.

The broader lesson here is one the industry keeps relearning: patching closes the vulnerability, but it doesn't retroactively invalidate credentials that were exfiltrated before the patch landed. For defenders, that gap — between "we patched" and "we're safe" — is exactly where attackers live.

There's also a market signal worth noting. FortiBleed dropped in the same week that three separate FortiSandbox CVEs entered active exploitation. That's not a coincidence — it reflects a sustained, multi-actor focus on Fortinet's product family. If your network boundary runs on Fortinet, this week is the week to audit that assumption.

Also today

Cyberattack Hits Mackay Sugar Mid-Harvest — Crops Left in the Ground

Australian sugar producer Mackay Sugar confirmed a cyberattack that disrupted operations during the peak cane crushing season — the worst possible time for a mill to lose system availability. The attack forced the company to leave harvested cane in the paddock, with delays compounding as the crushing window is narrow and weather-dependent. Details on the nature of the attack haven't been released, but the timing strongly suggests ransomware or a destructive payload targeting operational technology. It's a sharp reminder that OT environments in Australian agriculture carry genuine critical infrastructure risk, even when they fall outside formal SOCI Act critical infrastructure designations.

The Register

144 Mastra npm Packages Compromised in AI-Framework Supply Chain Attack

Researchers from five separate security firms jointly disclosed a software supply chain attack — codenamed "easy-day-js" — that compromised 144 npm packages in the @mastra namespace, a widely used JavaScript/TypeScript framework for building AI applications. The vector was a single hijacked contributor account, which gave attackers write access to the entire package namespace. Malicious versions were published before the compromise was detected. Any developer who pulled @mastra packages in the attack window should audit their dependency tree and rotate any secrets that ran in affected environments. The Mastra framework has significant traction in Australian developer communities building AI-adjacent tooling.

The Hacker News

Malicious JetBrains Marketplace Plugins Caught Stealing AI API Keys

At least 15 fake plugins on the JetBrains Marketplace were found exfiltrating AI API keys from developers, researchers confirmed. The plugins masqueraded as AI coding assistants built on DeepSeek and other popular LLMs — offering features like code review, commit message generation, and unit tests to lure downloads. Once installed, they quietly siphoned API keys for OpenAI, Anthropic, and other providers. The campaign appears coordinated, with plugins sharing common infrastructure. Developers using JetBrains IDEs — IntelliJ, PyCharm, WebStorm — should audit installed plugins and rotate any AI provider keys stored in their environments.

The Hacker News

RoguePlanet: Microsoft Defender Zero-Day Gets a CVE, Still No Patch

Microsoft has formally assigned CVE-2026-50656 (CVSS 7.8) to the Defender privilege escalation zero-day nicknamed "RoguePlanet," disclosed publicly a week ago — but a patch still isn't available. The flaw exploits a race condition in Microsoft's Malware Protection Engine to spawn a command prompt with System-level privileges. A working proof-of-concept has been public since disclosure, meaning any attacker with local access to a Windows machine running Defender can attempt the exploit right now. Microsoft says a fix is in development. In the meantime, organisations should consider tightening local access controls and monitoring for unusual process spawning from Defender-related executables.

SecurityWeek

G7 Leaders Agree on "Trusted Partners" AI Framework — With Sovereignty in Mind

G7 leaders concluded their Evian summit with a commitment to build out a "trusted partners" scheme for AI cooperation — diplomatic language for trying to reduce dependence on any single nation's AI infrastructure. The talks were shaped in part by the Anthropic model shutdown earlier this week, which rattled allies who discovered the US could cut off access to frontier AI overnight. French President Macron and Indian PM Modi were both vocal about the risk. Australia, as a G7 partner nation and Five Eyes member, is directly in scope for these frameworks — and the government's AI-in-government agenda is increasingly exposed to exactly this kind of unilateral supplier risk.

iTnews

World Leaders Want American AI — Just Not American Control Over It

The Anthropic blackout appears to have triggered a genuine foreign-policy reckoning. Reporting from TechCrunch notes that Macron and Modi used the G7 sidelines to raise explicit concerns that Washington could terminate access to American AI models at will — and that the Fable/Mythos shutdown proved the fear is not hypothetical. The geopolitical implication is that allied nations are now quietly accelerating domestic and European AI capability investments as a hedge. For AI vendors selling into government markets globally — including the growing cohort of Australian AI companies — "sovereign AI" is becoming a procurement requirement, not a marketing line.

TechCrunch

OpenAI's Near-Autonomous AI Chemist Improves a Real Drug-Making Reaction

OpenAI and drug-discovery startup Molecule.one published results showing a near-autonomous AI chemist — built on GPT-5.4 — successfully improved a challenging reaction in medicinal chemistry with minimal human intervention. The system autonomously designed experiments, interpreted results, and iterated on its approach. It's an early but concrete demonstration of AI moving from "helps scientists" to "acts like a scientist." The implications for pharmaceutical R&D timelines are significant: Australia's life sciences sector, including CSIRO and university-linked drug discovery labs, is watching closely as this capability class matures.

OpenAI Blog

Tesco Is Moving 40,000 Server Workloads Off VMware After 175% Price Hike

UK retail giant Tesco filed court documents claiming Broadcom hiked its VMware licensing costs by approximately 175 per cent following the acquisition — and is now actively migrating 40,000 server workloads to alternative virtualisation platforms. It's the most concrete public data point yet on what Broadcom's VMware repricing looks like at enterprise scale. Australian organisations across retail, financial services, and government are similarly locked into VMware estates, and many are navigating the same decision. HPE this week offered a year of free virtualisation software to tempt VMware customers — a sign the competitive window is open.

Ars Technica

Snap's $2,195 AR Glasses Arrive — and the Market Is Unimpressed

Snap debuted its long-awaited Specs AR glasses at $2,195, positioning them as a serious consumer computing device after more than 12 years of development. CEO Evan Spiegel described the product as bringing "computing into the world." Wall Street's verdict was swift: the stock fell sharply after the announcement. At more than double the price of Meta's Ray-Ban smart glasses, Specs faces an uphill battle convincing buyers the premium is warranted. The AR glasses market remains a graveyard of ambitious hardware plays — from Google Glass to Magic Leap — and the question of whether Snap has found the formula remains unanswered.

TechCrunch

Secure Boot Key Expiry Is Coming — Here's What Windows and Linux Users Need to Know

A deadline is approaching for updating Secure Boot keys on Windows and Linux systems, and it's getting less attention than it deserves. Secure Boot keys underpin the chain of trust that validates your machine's boot sequence — if they expire and aren't rotated, systems may fail to boot or lose boot integrity verification. The issue affects both consumer and enterprise machines, and the Linux community in particular has a patchwork of distributions with varying update cadences. Ars Technica has a clear breakdown of what the expiry means and what action is required. IT teams managing mixed Windows/Linux fleets should review their exposure now rather than scrambling after deadline.

Ars Technica

FIFA World Cup Streaming Had a Critical Access Control Flaw — Cameras Included

A security researcher disclosed an access control vulnerability in the FIFA World Cup streaming infrastructure that left live match streams wide open — and went further than just video. The flaw reportedly allowed the researcher to access broadcast camera controls and capture live stream keys, which would have enabled hijacking or disrupting live feeds. The vulnerability has since been fixed, but the disclosure highlights how major sporting events are increasingly complex attack surfaces, combining broadcast technology, third-party vendors, and real-time infrastructure under enormous public scrutiny. With Australia co-hosting the 2034 FIFA World Cup, this is worth filing away.

iTnews

UK's NCSC Chief: Three-Quarters of Critical Infrastructure Attacks Come From Hostile States

NCSC CEO Richard Horne told a RUSI conference that approximately 75 per cent of cyberattacks against Britain's critical infrastructure now originate from nation-state actors or their proxies — and that adversaries are "prepositioning" inside networks today for potential kinetic conflict tomorrow. Horne's framing was direct: intelligence gathered through cyber intrusions will guide physical targeting in any future conflict. The warning echoes posture assessments from Australia's ASD, which has similarly flagged pre-positioning activity in Australian critical infrastructure. The SOCI Act's notification and resilience obligations exist precisely to address this class of long-dwell threat.

The Record

Sources consulted