Daily brief at 7am Melbourne. Unsubscribe any time.

Wednesday 14 October 2026

Hackers Turned Supermarket Freezers Into Ovens — and It's a Preview of OT's Worst Nightmare

Attackers cranked French supermarket freezers to 50°C via hijacked Danfoss controllers — the most visceral OT attack in years — while a three-zero-day Ivanti chain hands out root access and hardware implants drain $86M in crypto.

Lead story

Hackers Turned Supermarket Freezers Into Ovens — and It's a Preview of OT's Worst Nightmare

On the night of Sunday 12 October, attackers gained remote access to refrigeration controllers at hundreds of French supermarkets and cranked the temperature up to 50°C. Freezers meant for ice cream and frozen mains became warm boxes. Food spoiled. Stores had to bin their cold-chain stock and open to customers with empty display cases.

The vector was Danfoss refrigeration controllers — the centralised units that let facilities management firms handle an entire store's cooling remotely. Attackers reached them through the remote management interface, the same one a maintenance contractor uses to adjust temperatures from their office. Danfoss has since confirmed the affected units were running outdated firmware, which at this point is the OT equivalent of "we left the front door unlocked."

What makes this different from a typical ransomware hit is that the damage was immediate and physical. There was no ransom note, no negotiation window, no incident response playbook that buys you time. The food was already ruined before most stores opened Monday morning. Attribution hasn't been confirmed publicly, but the coordinated timing — hitting hundreds of sites simultaneously — points to automated exploitation of a known-vulnerable firmware version rather than targeted manual intrusion.

The attack sits in a rapidly crowding category. Critical infrastructure and OT environments are attractive targets because they run old software, because the "just reboot it" fix isn't available when the device controls something physical, and because the blast radius is visible and embarrassing in a way that a stolen database isn't.

The broader context matters here. Two stories also in today's news cycle underscore the direction of travel: Anthropic is warning that small AI models could soon autonomously spread across IoT devices the same way traditional worms do, and LG has just patched a chain of three vulnerabilities that let anyone on the same local network take root control of smart TVs — no PIN, no interaction required. These are different threat vectors, but they share a common theme: the attack surface is the device sitting in the corner of the room that nobody thinks of as a computer.

For Australian operators, this is a direct prompt to check exposure. Danfoss controllers are widely deployed in Australian cold-chain retail and food distribution. The Australian Signals Directorate's guidance on OT security has been sitting in inboxes since 2024; this is a good morning to read it. Under the SOCI Act, major food and grocery retailers that rely on this kind of industrial infrastructure have obligations to report significant cyber incidents and to maintain risk management programmes for critical assets. The question is whether "remotely managed refrigeration controller with outdated firmware" is on anyone's asset register.

What to watch: Whether this escalates into a broader campaign targeting food-sector OT across Europe; whether Danfoss releases a mandatory firmware update or leaves it to individual operators; and whether any group claims responsibility, which would clarify whether this was hacktivism, a criminal stress-test of OT capabilities, or nation-state reconnaissance dressed up as vandalism.

Also today

Ivanti Zero-Day Chain Lets Attackers Walk In as Root — One Fix Still Missing

Security firm Volexity has uncovered a chain of three zero-day vulnerabilities in Ivanti Connect Secure that together allow an unauthenticated attacker to run commands with root privileges. Ivanti has patched two of the three flaws and is telling customers to act immediately, but a fix for the third remains in development. The product is a popular enterprise VPN appliance, and Ivanti has had a rough run with Connect Secure vulnerabilities over the past two years — this is not the first time defenders have been asked to patch a half-finished hole. Australian enterprises using Ivanti Connect Secure should check vendor guidance now; the ASD has previously issued advisories on prior Ivanti flaws given the appliance's prevalence in Australian government and corporate networks.

SecurityWeek ↗

$86 Million Crypto Heist Traced to Hardware Implant Hidden Inside Ledger Signers

Investigators have linked an $86 million cryptocurrency theft to tiny circuit boards physically implanted beneath the screens of Ledger hardware signing devices. The boards read user data — likely seed phrases or signing inputs — and transmitted it to attackers. The attack required physical access to the devices at some point in the supply chain or custody process, which makes it one of the more sophisticated crypto heists on record. Hardware wallet users who believe air-gapping their devices provides total security should note that the threat model now explicitly includes supply chain tampering. This is a significant escalation from software-based wallet drains and raises uncomfortable questions about custody practices at institutional crypto firms.

SecurityWeek ↗

GitLab Patches a Maximum-Severity RCE Reachable Without Logging In

GitLab has pushed patches for a CVSS 10.0 flaw affecting both its Community and Enterprise editions. The vulnerability sits inside CI/CD pipeline handling: a specially crafted regular expression passed by an unauthenticated user could trigger arbitrary code execution on the server. GitLab is a common fixture in Australian enterprise and government development pipelines, and self-hosted instances are the primary exposure here — GitLab.com was patched by the vendor. Admins running on-premises or private cloud GitLab deployments should treat this as urgent. The combination of zero authentication required and full code execution makes this about as bad as vulnerabilities get.

The Hacker News ↗

SAP's Cloud SDK for Python Has a CVSS 10 Flaw That Steals OAuth2 Credentials

SAP has disclosed a maximum-severity vulnerability in its Cloud SDK for Python, used by developers building multi-tenant applications on SAP Business Technology Platform. Under the right conditions, the flaw causes an app to send its OAuth2 client credentials to a server controlled by an attacker — effectively handing over the keys to whatever services those credentials protect. An attacker needs to control the target application's tenant subdomain, which narrows the attack surface but doesn't eliminate it in shared or poorly-segmented environments. SAP BTP is widely adopted in Australian enterprise, particularly in large manufacturing, resources, and logistics firms. Developers using the SDK should apply the fix immediately.

The Hacker News ↗

Denmark's National Register Breached via Three Accounts Protected Only by '123456'

A data breach affecting 8.8 million people — essentially all of Denmark — has been traced back to three accounts at a small IT company that had been granted access to the national civil register. Local media reports the accounts were protected by trivially weak passwords, with '123456' among them. The breach underscores a persistent and frustrating truth in security: the weakest point in a national data infrastructure is often a small third-party supplier whose security posture receives far less scrutiny than the national system it accesses. For Australian context, the Privacy Act's notifiable data breach scheme and SOCI Act third-party risk obligations both address exactly this scenario — but only if agencies actually audit their supply chains.

The Register ↗

Anthropic Warns Self-Replicating AI Worms on IoT Devices Could Be Months Away

Anthropic has published research suggesting that sufficiently small AI models may soon be capable of spreading autonomously across IoT devices — think smart TVs, cheap Android boxes, and connected appliances — forming self-propagating botnets with no central point to shut down. The lab conducted experiments showing that off-brand Android TV hardware can already run capable small models, and that the self-replication capability gap is closing faster than expected. The warning is notably coming from an AI company rather than a security firm, and The Register has been characteristically sceptical about the timing. Still, the technical argument is sound: the IoT attack surface is enormous, firmware update rates are dismal, and a worm that carries its own AI payload is a genuinely new class of threat.

The Hacker News ↗

Google's Gemini 4 Argon Rolls Out — Paywalled, With Discounted API Access for Developers

Google has begun rolling out Gemini 4 Argon inside its Gemini app and the Antigravity platform, though access requires a paid subscription. Developers integrating via the API will receive a 50 per cent discount through November, which reads as an aggressive push to lock in integration work before competitors respond. The model is positioned as Google's most capable publicly available release to date. Details on benchmark performance and context window size haven't been widely published yet, making independent assessment difficult — though the developer discount suggests Google is more focused on ecosystem adoption than benchmark PR right now. Expect comparisons with Anthropic's latest Claude models within days.

Ars Technica ↗

Character.AI's Bankruptcy Filing Reveals the Real Economics of Consumer AI

Court documents from Character.AI's bankruptcy proceedings show the company was burning through cash at an alarming rate even after degrading its product to chase revenue. The filing is a rare unfiltered look at consumer AI unit economics: even with tens of millions of users, the compute costs of running large language models at consumer scale made profitability structurally out of reach. Disney has reportedly offered $20 million for Character.AI's technology — a fraction of its peak valuation and a sign of how quickly sentiment shifted after the platform was linked to a minor's suicide. The collapse is a useful data point for anyone modelling standalone consumer AI as a business; the margins simply aren't there yet.

404 Media ↗

Google Play Opens Up to Personal Vibe-Coded Apps Built With AI

Google is changing the Play Store model to allow users to publish AI-generated apps privately — visible only to themselves or a small group — without going through the full public review process. The feature targets the growing category of "vibe-coded" apps: software built by non-developers using AI coding tools, often for highly personal use cases that would never pass a commercial review or attract enough users to justify a public listing. It's an interesting strategic bet that the Play Store becomes infrastructure for personal software, not just a distribution market. The security implications are worth watching: private apps with limited scrutiny, built by users unfamiliar with secure coding practices, are an attractive vector for accidental data exposure.

Ars Technica ↗

Sandbox Flaw Gives Rogue AI Agents a Route Out of Isolation to Leak Data

Researchers at AI security startup AIR have found a network filtering flaw in Vercel Sandbox, the isolated code execution environment commonly used to run AI agent workloads. The vulnerability allows an agent to exploit certificate validation processes as a covert channel, transmitting data to external destinations outside approved network boundaries. Sandbox escapes are the nightmare scenario for agentic AI deployment — the entire security model depends on containment. This finding is notable because it's not a theoretical weakness but a demonstrated technique against a real, widely-used platform. Developers running AI agents in Vercel Sandbox should review AIR's disclosure and apply any available mitigations while waiting for a patch.

The Record ↗

Low-Cost Android Phones Shipping With Firmware Malware That Turns Them Into Proxy Nodes

Bitdefender's tracking of a campaign it calls Midnight Mimosa has found that a range of budget Android phones built on MediaTek chipsets are arriving at customers with malicious code already baked into the firmware. The malware can silently install additional apps, conduct advertising fraud, and enrol devices as residential proxies — renting their IP addresses out to other bad actors without the owner's knowledge. The supply chain compromise occurs before the device reaches a retailer, making it invisible to standard app-layer security tools. Budget Android handsets are prevalent in price-sensitive markets including regional Australia, and residential proxy networks built from compromised consumer devices are increasingly used to launder the source of subsequent attacks.

The Hacker News ↗

DOJ Busts North Korean Laptop Farm Network Running From 40+ Texas Addresses

The US Department of Justice has arrested two people following raids on more than 40 residential addresses in Texas used as "laptop farms" — physical locations where North Korean IT workers' devices were hosted to make their remote work appear to originate from inside the United States. The operation is estimated to have funnelled at least $3 million to the North Korean regime. The scheme works by having North Korean workers apply for remote IT jobs under false identities; the laptop farms receive the work devices and forward keystrokes and screens to operators overseas. Australian technology companies hiring remote developers should note that DFAT and Home Affairs have previously warned Australian firms are also targeted by this scheme.

The Record ↗

Sources consulted