Allan Grant | SOC Analyst | 28 October 2026
Attackers put AI to work. Attackers exploited two decade-old flaws in PHP, which runs most of the web, before the fix was released, and researchers used AI to help turn the flaws into working attacks. Criminals pointed autonomous AI agents at more than 1,200 websites, and a hacker used ARTEX, a Chinese agentic tool, to steal data from South Korean banks.
Cyber attacks caused physical damage. Attackers took over refrigeration controllers in hundreds of French supermarkets and set them to defrost overnight, spoiling stock and starting fires.
ASD replaced the Essential Eight. ASD published ASD Essentials as the successor framework, and its Annual Cyber Threat Report named AI the defining force of the year.
Suppliers were the way in. Many Australian incidents in October, as with previous months, came through a third party, from booking software to court transcription contractors.
ShinyHunters is the extortion group behind several of 2026's largest data thefts, including the Instructure and Oracle PeopleSoft campaigns covered in earlier updates. Its May breach of Instructure's Canvas platform reached Queensland's Department of Education and dozens of other Australian education bodies.
In September, ShinyHunters claimed to have breached the FBI's own recruitment site and stolen data on almost all FBI agents and applicants. The FBI and its partners answered with at least three arrests in early October, including the group's alleged leader in Jordan. The FBI claims that person took part in hacks on more than 140 organisations and collected at least US$70 million in extortion payments (SecurityWeek).
The investigation has now implicated the people victims hire to negotiate. On 8 October the FBI arrested Edward Dubrovsky, a former executive at Canadian negotiation firm CYPFER, on extortion charges. Investigators allege he knew of ShinyHunters victims before they did, and suspect an executive at a US consulting firm helped run the group's extortion site (KrebsOnSecurity).
Some negotiators take a percentage of the ransom, which rewards them when victims pay more. The FBI has advised victims not to pay ShinyHunters, and responders warn that data the group promises to delete is often kept, resold or used for later extortion.
On 8 October Australia joined the US, UK, Canada, Japan, New Zealand and Spain in naming Integrity Technology Group, a Beijing company with Chinese government links, as the engine behind Flax Typhoon. That state-sponsored group has targeted critical infrastructure, government and telecommunications since at least 2021. The firm builds hacking tools, runs attack infrastructure and pairs large-scale automated scanning with hands-on intrusions to steal sensitive data. US authorities seized two of its tools the same day (Cyber.gov.au; Help Net Security).
On 14 October the US Federal Trade Commission sent civil subpoenas to OpenAI and Anthropic seeking records on incidents where their AI agents accessed corporate and government systems without authorisation. The FTC also plans to compel executives from both companies and the independent evaluator METR to testify (The Hill; CDO Magazine). The incidents behind the probe include OpenAI agents breaching several Australian government websites, covered in this and last month's updates.
On the night of 11 October, attackers took over internet-connected refrigeration controllers in hundreds of French supermarkets, including Intermarché, Carrefour and Monoprix stores, and set them to defrost. Fridges reached about 50°C, stores threw out their stock, some caught fire, and one chain estimated its losses at several million euros. Stores in at least five other European countries were reportedly hit (BleepingComputer; CBS News; TIME).
Many supermarkets run their cold rooms from panels reachable over the internet, so maintenance contractors can adjust them off-site. Investigators have not confirmed the exact entry point, but Danfoss said most affected units were end-of-life models that no longer receive security updates, and newer units at some stores were running old firmware. Researchers had also documented a login bypass that let anyone reaching the management page generate a valid access code from the device's date (BankInfoSecurity; Nozomi Networks).
Organisations that rely on remotely managed operational technology should know what is connected to the internet, who can reach it, and whether it still receives security updates.
A hacktivist group calling itself the Department of Peace leaked the names, contact details and roles of 41,191 Department of Homeland Security staff, with Customs and Border Protection and ICE the largest groups, along with more than 55,000 internal group names that reveal undercover units. The transparency collective DDoSecrets published the dataset on 28 October and says it came from the Okta single sign-on system DHS uses. Okta says its own systems were not compromised, and details of several staff named in the leak have been independently verified (Help Net Security; CNBC).
From late September to early October, a single Chinese-speaking attacker stole customer data from several South Korean financial institutions, including Shinhan Bank, KB Kookmin Bank and Hana Bank, exposing personal and credit card information and causing outages at some. The attacker used ARTEX, an open-source AI penetration-testing tool developed in China, alongside Claude Code and other AI models to automate reconnaissance and exploitation. The attacker then asked Claude where to sell the stolen Korean data. South Korea's government held an emergency meeting, and financial regulators warned customers to expect phishing and loan scams using the leaked data (BleepingComputer; The Hacker News).
One financially motivated individual with the right tools can now run several intrusions against well-defended targets in a matter of weeks. Plan for that as the baseline threat.
PHP is the programming language behind an estimated 70% of all websites, including those built on WordPress, Drupal and Magento, and it also runs inside many vendor products and appliances. In October researchers showed that two flaws, dubbed PHP2Shell, let an attacker take over a server without logging in. The bugs were first found in 2024 but were not thought exploitable and went unpatched. Researchers at Calif said AI helped them build a working exploit, though most of the work drew on their own expertise. The flaws affect every PHP version back to 5.4, released in 2012, and because they are in PHP itself, a site was exposed even if its own code was sound (TechCrunch).
Defenders had little warning. The two CVE identifiers began circulating online on 20 October, and PHP released the fix on 21 October, a day earlier than planned, after one vendor published patched images more than six hours before the coordinated release. Exploitation attempts date back to 17 October, and government agencies and companies worldwide took websites offline while they patched. On 23 October the PHP Security Team disclosed a third flaw that the first fix did not address, although exploiting it requires an uncommon combination of PHP configuration and application behaviour (CSO Online; Canadian Centre for Cyber Security).
In the most serious confirmed case, a suspected China-linked actor scanned about 19,000 internet-facing Palo Alto Networks firewalls and chained PHP2Shell with a Linux kernel flaw, Fragnesia, to take full control of 24 of them and an internal firewall management server across the US, Japan, Taiwan and South Korea, including systems in a US government-affiliated network, before going after internal login systems. AI tools were found on the attacker's infrastructure (The Register; Dream).
In PHP2Shell the month's two trends met: AI is helping turn old, overlooked flaws in widely shared software into working attacks, and attackers are exploiting them faster than organisations can patch. A flaw in a web language became a way through security appliances, so inventories that stop at web servers will miss it. The government response, including the ACSC's critical alert, is covered under Significant Advisories and Vulnerabilities below.
On 15 October ASD published ASD Essentials, with chapters for enterprise IT, operational technology and cloud, plus AI-focused guidance. ASD also released a transition guide, an implementation planning guide, an assessment process guide and a senior executive briefing. Assessors now rate each control on how well it works, replacing Maturity Levels One to Three, and existing maturity ratings will not carry across (ASD).
ASD has not set a retirement date for the Essential Eight.
ASD released its seventh Annual Cyber Threat Report on 27 October and called AI the defining force in cyber threats over the past year (Cyber.gov.au). In 2025–26 ASD's ACSC:
Small businesses reported an average cost of $55,400 per cybercrime, medium businesses $141,300 (up 45%) and large businesses $148,100 (business.gov.au). ASD found attackers exploit most vulnerabilities within two weeks, some within 48 hours, and expects AI to shrink that window.
The OAIC's Notifiable Data Breaches statistics for January to June 2026, released on 26 October, show 667 notifications, only slightly up from 662. The size of breaches grew sharply: breaches affecting 500,000 to 1,000,000 people rose from none to 3, and those reported to the OAIC affecting more than 1,000,000 people globally rose from 4 to 11. Malicious attacks remain the leading cause, and health was the most targeted sector. Organisations are detecting breaches faster, with more than half identified within 10 days (OAIC).
Privacy Commissioner Carly Kind pointed to data minimisation: organisations cannot lose personal information they do not hold.
On 23 October Australia imposed financial sanctions and travel bans on two people and two companies that help North Korea place fraudulent IT workers in foreign businesses to fund its weapons programs (Minister for Foreign Affairs). The listings act on the Multilateral Sanctions Monitoring Team report covered in last month's update. They formed part of a month of coordinated action: the EU sanctioned North Korean worker deployments to Russia and an arms-for-oil trade on 12 October, the US issued new sanctions, and on 28 October 11 countries and the EU, including Australia, the US, the UK and Japan, backed the measures in a joint statement (UPI via Yahoo News). These workers win remote roles by posing as citizens of other countries, so the risk sits in hiring and contractor onboarding as much as in security controls.
On 22 October Australia and Timor-Leste signed a Defence Cooperation Framework backed by a $62 million package over five years. More than $19 million goes to cyber security: specialist advisers, incident response support and protection of critical systems, plus digital forensics to help Timor-Leste's police investigate cybercrime and disrupt scam centres (TATOLI; South China Morning Post). The deal follows similar agreements with Vanuatu, Fiji and Papua New Guinea as Canberra works to limit China's security role in the region, and confirms cyber security as a standard part of Australia's regional security deals.
Australian organisations disclosed incidents in October across tourism, retail, health, software, recruitment, education, hospitality, manufacturing, media and emergency services. As in August and September, attackers went through a third-party provider in most cases.
About 1 October, OpenAI told the NSW Government that one of its agents had accessed non-public historical fire statistics from the NSW National Parks and Wildlife Service in June. This was the fifth Australian government system OpenAI has disclosed, after Medicare statistics, the AIHW, the Victorian Department of Health and the NSW Bureau of Crime Statistics and Research (ABC News; Digital Trends).
The incidents are now shaping policy. At the Joint Select Committee on Artificial Intelligence's 6 October hearing in Sydney, OpenAI and Anthropic both said they would support mandatory reporting of breaches involving AI agents, and OpenAI acknowledged it should have told the government sooner about the Medicare portal access (Quartz; Digital Watch). On 12 October the committee sent early advice to the Prime Minister recommending independent safety evaluators inside frontier AI labs with a duty to report incidents, reserved compute capacity for national interest uses, and early access to new models for security agencies such as ASD as a condition of training models in Australia. Its final report is due by 30 November (ABC News; Parliament of Australia).
The PHP2Shell flaws described under International Incidents above led ASD's ACSC to issue a critical alert on 21 October, before full details were public, warning that all versions from PHP 5 onwards may be affected. CISA confirmed active exploitation on 26 October, and on 23 October the PHP Security Team disclosed a third, harder-to-exploit flaw that the first fix does not address (Cyber.gov.au; CISA; Canadian Centre for Cyber Security).
Few organisations can say where PHP runs across their own sites, vendor products and hosted services. Executives should ask for that inventory and confirmation that every instance is patched.
On 23 October the ACSC urged all Fortinet customers to apply updates for critical vulnerabilities across a wide range of products. The ACSC knows of no exploitation in Australia and expects Fortinet to publish more detail on 10 November (Cyber.gov.au).
In early October the ACSC updated September's NetScaler critical alert after Australian organisations confirmed exploitation, and recommended reviewing for compromise since at least 4 September. On 9 October it added a further critical flaw affecting NetScaler appliances configured for SAML single sign-on (Cyber.gov.au).
On 14 October, days after the French supermarket attack, the ACSC issued a high alert on cyber incidents affecting commercial refrigeration systems overseas. It has not seen active targeting in Australia, but urged organisations running internet-connected operational technology, including refrigeration, to assess their exposure and follow its guidance on OT environments and securing edge devices (Cyber.gov.au).
In October attackers used AI to move faster, but they still got in through old gaps: unpatched servers, exposed controllers and over-trusted suppliers. Organisations that patch fast and know who owns each asset and supplier will be well placed as ASD Essentials beds in.
Request a consultation with one of our security specialists today or sign up to receive our monthly newsletter via email.
Get in touch Sign up!
Fortian
Specialist Australian cyber security. Consulting, 24x7 managed security and privacy services since 2013.
1800 717 545