Wednesday 10 June 2026

Patch Tuesday from Hell: 206 Microsoft Fixes, a Chrome Zero-Day, and a Veeam RCE Land on the Same Day

Microsoft's record-breaking 206-vulnerability Patch Tuesday collides with a Chrome zero-day and a Veeam RCE flaw — defenders, your patching queue just got very long.

Lead story

Patch Tuesday from Hell: 206 Microsoft Fixes, a Chrome Zero-Day, and a Veeam RCE Land on the Same Day

Yesterday was one of the busiest patch days on record. Microsoft dropped fixes for 206 vulnerabilities — the largest Patch Tuesday haul the company has ever shipped — including three zero-days that were publicly known before a patch existed. That last part matters: public disclosure without a patch is an open invitation, and defenders were essentially running blind on at least some of these for a window of time.

The three zero-days span a range of severity. One involves a Windows privilege escalation flaw that was already being discussed in the wild. A separate disclosure that got its own headlines: a researcher known as Nightmare Eclipse had a notably public feud with Microsoft about responsible disclosure timelines, and the patch for at least one of their reported flaws appears to have landed in this batch. The dynamic — researcher going loud, vendor scrambling — is a reminder that the coordination between finders and fixers is still far from smooth.

But Microsoft wasn't the only one shipping urgent fixes. Google issued an emergency update for Chrome addressing CVE-2026-11645, an out-of-bounds memory access bug in V8 — Chrome's JavaScript engine — that's already being exploited in the wild. This is Chrome's fifth exploited zero-day of 2026. Google paid the discovering researcher a $55,000 bounty. The fix is in Chrome 149.0.7827.103; if you haven't restarted your browser recently, you're probably still running a vulnerable version. Chrome is ubiquitous across Australian enterprise desktops and consumer devices alike.

Veeam also joined the party, patching CVE-2026-44963, a remote code execution flaw in Backup & Replication with a CVSS score of 9.4. Any authenticated domain user could exploit it to run arbitrary code on the backup server. Backup servers are crown-jewel infrastructure — they're what organisations reach for when ransomware hits. A compromised backup server turns a bad day into a catastrophic one. Veeam has been a recurring ransomware target for years; patches here should be treated as drop-everything urgent. Australian organisations using Veeam in domain-joined environments — extremely common across enterprise and government — should prioritise this immediately.

SAP also patched 15 vulnerabilities, four of them critical, affecting NetWeaver and Commerce Cloud. NetWeaver in particular has been a persistent target for nation-state actors this year, and SAP environments tend to sit at the heart of enterprise finance and supply chain operations.

Adobe rounded out patch day with fixes for 123 vulnerabilities across its portfolio, nearly half of them in Experience Manager — Adobe's enterprise web content platform. Arbitrary code execution bugs in a platform that sits on public-facing web infrastructure deserve immediate attention.

The volume here is genuinely worrying in aggregate. CyberScoop noted fears about a "roaring flood of error-riddled software" materialising — and this patch cycle looks like evidence of that. The same AI-assisted vulnerability discovery tools that help defenders find bugs faster are also accelerating the pace at which offensive researchers and threat actors locate them. The patch window — the time between a fix being available and attackers weaponising the vulnerability — is shrinking.

What to do: Prioritise the Chrome update (browser restart, now), Veeam Backup & Replication (patch before the weekend), and the three Microsoft zero-days. SAP NetWeaver follows closely. If your organisation uses Check Point VPN products, CISA has given US federal agencies just 72 hours to patch a separate authentication bypass being actively used by Qilin ransomware affiliates — treat that timeline as a reasonable benchmark regardless of jurisdiction.

Also today

Anthropic Releases Claude Fable 5 — Mythos Power, Guardrails Included

Anthropic has publicly released Claude Fable 5, its first Mythos-class model available to everyone. Fable 5 sits just below the full Mythos model — which remains restricted to vetted partners via Project Glasswing — but Anthropic says it outperforms previous models on software engineering, knowledge work, and vision tasks, with its edge growing on longer, more complex jobs. The catch: it refuses to engage with cybersecurity exploits, certain biological queries, and chemistry topics Anthropic deems high-risk. Whether those guardrails hold under adversarial prompting remains to be seen. Simultaneously, Anthropic quietly upgraded Glasswing partners to Mythos 5. Both releases are available via API and the Claude.ai interface.

WIRED Security ↗

Claude Mythos Turns N-Days Into Exploits in Hours — and That's a Problem for Everyone

New testing by XBOW confirms what security researchers have been quietly worried about: Anthropic's Mythos model can take a known, patched vulnerability and build a working exploit far faster than a human researcher — compressing what used to take days into hours. The implication is stark. The patch window — already shrinking — could collapse further if adversaries gain access to Mythos-class capabilities without the guardrails. SecurityWeek notes that even public models with safety filters disabled are showing similar acceleration. The bug bounty industry faces an existential rethink: if finding vulnerabilities becomes trivially cheap, the economics of the whole ecosystem shift.

SecurityWeek ↗

France's Government Messaging App Breached via Hijacked Account

DINUM, France's digital government directorate, confirmed that Tchap — the encrypted messaging platform used by French public servants — was accessed by attackers who hijacked a legitimate user account. The agency says only public chat rooms were exposed, but a person claiming responsibility alleges far deeper access. Tchap was built as a sovereign alternative to commercial messaging tools, running on the open-source Matrix protocol. The breach is a pointed reminder that even purpose-built secure platforms are only as strong as their account security. Australia's whole-of-government messaging posture — still largely reliant on commercial platforms — faces similar account-takeover risk vectors.

Bleeping Computer ↗

FROST Attack: Websites Can Fingerprint Your Browsing History Using SSD Timing

Researchers at Graz University of Technology have demonstrated a new side-channel attack called FROST that allows a malicious website to infer which other sites you've visited and which apps you've opened — using nothing but JavaScript and the timing of SSD read contention. No permissions, no browser extensions, no native code required. The attack works by watching how long it takes the drive to respond under load, effectively eavesdropping on background disk activity. It functions across common NVMe SSDs. The attack needs time to work, so short page visits reduce exposure, but a sitting browser tab is enough. Full technical details were published by the Graz team.

The Hacker News ↗

Russia-Aligned Groups Exploit Unpatched WinRAR Flaw Against Ukraine — a Year After the Fix Shipped

Two Russia-linked threat groups — Gamaredon and UAC-0226 — are actively exploiting CVE-2025-8088, a path traversal flaw in WinRAR, against Ukrainian military and government targets. The vulnerability was patched nearly a year ago. Trend Micro's attribution names both groups as distinct campaigns, though their tooling overlaps. The lesson here isn't new but keeps proving itself: patching velocity in conflict-zone IT environments is constrained by operational tempo, and adversaries absolutely know it. WinRAR remains extraordinarily widespread in Eastern European enterprise environments. Ukrainian defenders are effectively playing defence with a delayed bench.

The Hacker News ↗

Hades PyPI Attack: 19 Packages Poisoned in Supply Chain Wave Linked to Miasma

The Miasma supply chain campaign — which last week hit 73 Microsoft GitHub repositories — has spawned a new branch called Hades, targeting the Python Package Index. Researchers found 37 malicious wheel artifacts spread across 19 PyPI packages, each shipping a hidden setup file that executes automatically at install time and drops a Bun-based credential stealer. The attack is getting more modular: rather than one campaign hitting everything, the toolkit is being refactored for specific ecosystems. PyPI is heavily used across Australian fintech, academic, and government Python deployments. If your team pulls packages from PyPI without verifying checksums or using a private registry, now is the time to review that workflow.

The Hacker News ↗

Microsoft Exchange Flaw Lets Attackers Send Email as Anyone

A newly disclosed technique dubbed "Ghost-Sender" exploits a flaw in how Microsoft Exchange handles hybrid mail routing to allow attackers to spoof email from any address — including CEO, finance, or IT domains — in a way that bypasses standard authentication checks. The attack requires Exchange Online or on-premises Exchange running in hybrid mode alongside a third-party mail server or spam filter. It's an ideal setup for business email compromise at scale. Dark Reading reports the technique is already attracting attention in threat actor circles. Australian organisations running hybrid Exchange environments — still common across mid-market firms — should review their mail routing and DMARC enforcement configurations.

Dark Reading ↗

CISA Overhauls How It Ranks Vulnerabilities — a Binding Directive Is Coming

CISA acting director Nick Andersen announced the agency is fundamentally rethinking how it assesses and prioritises cyber risk for both federal agencies and critical infrastructure operators. A binding operational directive is due Wednesday that will require agencies to elevate some vulnerabilities while deprioritising others — moving away from treating all CVEs as roughly equal. The shift reflects a growing recognition that the current model, driven largely by CVSS scores, doesn't reflect actual exploitation likelihood. Australia's ACSC has been moving in a similar direction with its own risk-tiering guidance; local critical infrastructure operators covered by the SOCI Act should watch how this US framework evolves.

The Record ↗

Apple's WWDC AI Play: Automatic Password Fixes, Private Cloud AI, and an EU Standoff

Apple's WWDC 2026 keynote was heavy on AI — but understated about it. The headline features include an iOS 27 tool that automatically detects compromised passwords and changes them on your behalf via Safari, and a new Siri architecture that can run more complex queries through Google's cloud infrastructure while, Apple insists, keeping the data opaque to Google. The more politically interesting subplot: Apple is withholding the upgraded AI Siri from EU users, blaming DMA interoperability requirements, and openly asking European consumers to pressure their regulators. Australian iPhone users will get the new features; the broader question of whether Apple's privacy claims for its cloud AI hold up under scrutiny remains open.

TechCrunch AI ↗

Meta Is Now Using What You Do on Other Websites to Personalise Your Feed and AI

Meta announced it will expand its use of off-platform activity data — the behavioural signals businesses share about their customers via the Meta pixel and similar tools — to personalise not just ads but also your Reels feed and responses from Meta AI. So if you bought camping gear on another site, expect camping content in your Instagram feed. Meta frames this as improving relevance; critics frame it as extending surveillance capitalism into the AI layer. Separately, and quietly, facial recognition code was found in Meta's Ray-Ban smart glasses firmware and removed within 24 hours of discovery. The company has not explained what it was doing there. Australia's Privacy Act reforms, currently before parliament, directly address this kind of cross-context data use.

The Verge ↗

A Single Wrong Character in the Linux Kernel Unlocks Root Access

A high-severity use-after-free vulnerability in the Linux kernel — traced to a single errant character in the source code — can be exploited to escape sandbox defences and gain root privileges on affected systems. Ars Technica reports the bug affects the kernel's memory management subsystem and is reachable from within containers and restricted environments, making it particularly relevant to cloud and shared-hosting deployments. The flaw has been patched in mainline kernel releases, but distribution uptake varies. Given how broadly Linux underpins Australian government cloud infrastructure, managed hosting, and enterprise server fleets, kernel patch cadence deserves attention here.

Ars Technica ↗

Sources consulted