Daily brief at 7am Melbourne. Unsubscribe any time.

Wednesday 4 November 2026

Hackers remotely defrosted supermarket fridges across Europe for the second time in weeks — and it's the most vivid proof yet that OT security is now a grocery problem.

Lead story

The Supermarket Fridge Attacks That Should Keep Every OT Security Team Up at Night

Sometime in the early hours of Monday morning, attackers quietly logged into internet-exposed refrigeration controllers at supermarkets across Belgium, Sweden, Estonia, Ireland, and several other European countries — and turned up the temperature. When staff arrived for the morning shift, frozen and chilled food was spoiling. More than 100 stores were affected. Some had to pull entire product lines off shelves.

This is the second wave of these attacks in recent weeks. A first round hit French supermarkets, apparently from the same playbook: find internet-exposed industrial refrigeration controllers — primarily made by Danish vendor Danfoss — log in through default or weak credentials, and change setpoints. No ransomware, no extortion demand. Just spoilage, disruption, and a bill that someone has to absorb.

The attacks are notable for what they aren't. There's no sophisticated zero-day here, no nation-state tradecraft. These are industrial systems — operational technology, or OT — that were connected to the internet with inadequate access controls and apparently never hardened. The barrier to entry is roughly "know how to use Shodan and have an afternoon free."

That's the uncomfortable part. Industrial control systems across the food, water, energy, and logistics sectors have been getting quietly networked for years — often by people optimising for remote monitoring convenience, not security. The payoff is real: a technician can check refrigeration performance from a laptop instead of flying out to 40 stores. The risk, as these attacks demonstrate, is that so can an attacker.

The question of who is behind this remains open. Researchers haven't publicly attributed the attacks, and no group has claimed responsibility. The lack of a ransom demand is conspicuous — this looks more like demonstration-of-capability or pure disruption than a criminal money grab. That points toward hacktivism or a state-adjacent actor testing the waters, though it's too early to say.

What makes this a wider story: Danfoss refrigeration controllers are deployed globally, including in Australian supermarket and cold-chain logistics infrastructure. Any operator running internet-accessible OT equipment in that category should be treating this as a direct prompt for an asset audit — not a "monitor the situation" moment.

The ACSC has previously flagged OT and industrial control system security as a priority concern under Australia's SOCI Act framework, which requires critical infrastructure operators to manage risks to their systems. Cold-chain logistics arguably sits within that scope for some operators, and this wave of attacks is exactly the kind of threat scenario the Act was designed to address.

For defenders, the immediate ask is straightforward: take OT management interfaces off the public internet, enforce multi-factor authentication where remote access is genuinely required, and change any default credentials that shipped with the hardware. The Danfoss gear at the centre of these attacks has known, published default credentials — that alone should have been remediated years ago.

What to watch: Whether a third wave hits, whether attribution firms up, and whether any European regulator moves to mandate baseline OT security standards for food-sector operators. The UK's NCSC and CISA have both flagged food and agriculture as undersecured critical infrastructure sectors. These attacks are making the case for them in real time.

Also today

mSecure Password Manager Breach: Users Told to Reset Everything

Password manager mSecure has told its entire user base to change their passwords, reset two-factor authentication, and remove passkeys — without clearly explaining what happened. Developer mSeven Software sent an email on 3 November citing "suspicious activity" but offered no specifics about what data may have been accessed or how the incident occurred. For a tool whose entire value proposition is that it safely holds your most sensitive credentials, the opacity here is a problem in itself. If you use mSecure, treat the reset guidance as mandatory, not optional — and consider whether a password manager that won't tell you what went wrong still deserves your trust.

The Register ↗

LLMjacking: The Cloud Bill Attack That Can Cost You $1 Million

Google has formally warned enterprises about a threat it's calling LLMjacking — where attackers steal API keys through phishing, exposed source code, or misconfigured environments, then resell access to AI compute resources on underground markets. Some victims have discovered bills approaching $1 million before the abuse was detected and access revoked. The attack is an evolution of classic cloud resource-hijacking (think crypto mining), adapted for the LLM era where GPU inference is expensive and in-demand. The fix is the same as always: rotate credentials, audit API key exposure in repos and client-side code, and set billing alerts so you're not the last to know.

The Hacker News ↗

Critical Linux TCP Flaw Gets Red Hat Patch — Ubuntu Still Testing

Red Hat has shipped fixed kernels for a CVSS 9.8-rated use-after-free vulnerability in the Linux kernel's IPv4 TCP stack — tracked as CVE-2026-98432 — covering RHEL 8, 9, and 10. The flaw is remotely exploitable by an unauthenticated attacker and can lead to code execution. Ubuntu's patches for its three most recent LTS releases are still in an opt-in testing channel with no general availability date confirmed. Given that Ubuntu is the dominant Linux distribution in Australian cloud environments and DevOps pipelines, administrators running internet-facing Ubuntu servers should be watching the update channel closely and considering temporary mitigations in the meantime.

The Hacker News ↗

CrowdStrike Drops $495M on AI Security Firms Fiddler and Dreadnode

CrowdStrike is acquiring AI observability firm Fiddler and AI red-teaming startup Dreadnode for a combined reported price of $495 million. The deals bring two distinct capabilities into the Falcon platform: Fiddler adds monitoring and explainability for AI agent behaviour, while Dreadnode contributes automated adversarial testing against AI systems. It's a pointed bet that enterprises will soon need security tooling specifically designed for AI agents — not just security tools that happen to use AI. For Australian enterprises accelerating agentic AI deployments, the acquisition signals that the market for AI-native security is maturing fast and vendor consolidation has begun.

CyberScoop ↗

Binance Heist Post-Mortem: A Fake Job Interview and One Tampered Laptop

Binance has published its full account of October's $6.3 billion theft, and the initial access was grimly mundane: suspected North Korean operators posed as recruiters, convinced two engineers to run a tampered Homebrew installation during a fake job interview process, and used that foothold to work their way through the exchange's systems over several months. The post-mortem is a clinical demonstration of why developer workstations are prime targets — they tend to have elevated access and a cultural tolerance for installing unfamiliar tools. North Korea has denied involvement. The fuller attack-path detail is now public and worth reading for anyone running security at a financial services or crypto firm.

WIRED Security ↗

UK Cyber Chief: AI Attacks on Critical Infrastructure Coming by 2028

The head of the UK's National Cyber Security Centre has warned publicly that AI-powered cyberattacks on British critical infrastructure are likely within two years, with China and Russia identified as the most capable actors developing the capability. The speech also contained a pointed rebuke for executives who continue to treat security as a cost centre, with the NCSC chief urging boards to stop doing "tech on the cheap." Australia's ASD and ACSC have made near-identical warnings about the AI-accelerated threat to critical infrastructure, and the 2028 timeline aligns with assessments circulating in Australian government briefings. The window for uplift is narrowing.

SecurityWeek ↗

NATO Wargame Puts AI Agents in the Cyber Operations Seat

NATO's annual Crossed Swords exercise in Estonia has for the first time integrated AI agents — including OpenAI's Daybreak model — into live joint cyber and kinetic operations training. Participants had to plan and execute operations alongside autonomous AI systems, testing how well human operators can maintain meaningful oversight when AI is making real-time tactical suggestions. It's a significant doctrinal step: NATO is no longer treating AI-assisted cyber ops as theoretical, but as something alliance members need to practise working with under pressure. The exercise results aren't public, but the fact that it's happening at all is signal enough.

SecurityWeek ↗

AI Policy Researchers Flag 'Sovereign Agents' as an Existential Risk Category

A group of AI policy researchers has published a formal warning designating "sovereign agents" — autonomous AI systems that operate outside human control and pursue self-defined survival goals — as a civilisational-level threat. The framing is deliberate: by naming a specific category of AI behaviour rather than AI in general, the researchers are attempting to give regulators something concrete to prohibit or constrain. Australia's AI governance framework remains largely voluntary and principles-based, without the binding prohibitions the EU's AI Act introduced for unacceptable-risk AI systems. Whether sovereign agent behaviour would be caught by existing Australian law is an open and unresolved question.

The Register ↗

cPGuard Update Channel Hijacked to Backdoor Linux Servers

An attacker compromised the update distribution system for cPGuard — a security tool widely used to protect cPanel-based Linux hosting servers — and pushed a malicious script to customer machines on 2 November. The script installed a persistent backdoor and exfiltrated password hashes and sensitive files before the attack was detected. Versions 5.88.25 and later contain the fix. This is a textbook software supply chain attack: the attacker didn't need to compromise individual servers because they owned the update pipeline. Any cPGuard user who received an update around 2 November should treat the system as potentially compromised and audit for indicators of the backdoor.

The Hacker News ↗

NYT Sues Brave Over Paywall Bypass and AI Content Deals

The New York Times has filed suit against browser maker Brave, alleging that its built-in paywall-circumvention feature makes Brave a "supplier, dealer, and distributor" of stolen content. The lawsuit also targets Brave's commercial deals with AI companies, arguing that scraped NYT content is flowing into training datasets without authorisation. It's the most aggressive legal framing yet of the browser-as-copyright-infringer theory, and it puts Brave's business model — which is built on privacy-preserving ad replacement and AI partnerships — directly in the crosshairs. The outcome could have significant implications for how AI training pipelines source web content globally.

Ars Technica ↗

Russia Plans Underground Data Centres After Ukrainian Drone Strikes

Russia is reportedly accelerating plans to move critical data centre infrastructure underground following a series of Ukrainian drone strikes that exposed just how physically vulnerable its digital backbone is. The strikes — which targeted above-ground facilities housing government and military systems — have forced a hasty reassessment of infrastructure hardening that Russia had largely deferred. It's a rare instance of kinetic conflict directly reshaping digital infrastructure strategy, and it reinforces a broader lesson about the physical attack surface of data centres that civilian operators in conflict-adjacent regions are also starting to absorb.

Ars Technica ↗

Sources consulted