The NCSC says a China-linked hacking campaign uniquely used AI tools to steal data from critical sectors worldwide. Its own 58-page advisory with the FBI never says AI once, and blames a nine-year-old Python script.
Estimated reading time: 5 minutes
In short. On 8 October 2026, the National Cyber Security Centre announced a joint advisory with the FBI, CISA, NSA and five other international agencies about a China-linked hacking operation enabled by a company called Integrity Technology Group. The NCSC's own announcement says the actors are "uniquely using AI tools" to steal data from companies worldwide. The 58-page technical advisory the NCSC co-authored and links to as its evidence never once uses the words AI, artificial intelligence or machine learning, and describes the group's central tool as a Python-based penetration-testing script in use since at least 2017, built on vulnerabilities dating back to 2014. One documented mistake: four stars.
Eight security agencies from seven countries do not often put their names to the same document. On 8 October 2026 the FBI, the US Cybersecurity and Infrastructure Security Agency, the National Security Agency, the UK's National Cyber Security Centre, Australia's Cyber Security Centre, the Canadian Centre for Cyber Security, Japan's National Police Agency and National Cybersecurity Office, New Zealand's NCSC and Spain's Centro Nacional de Inteligencia all co-signed a technical advisory naming Integrity Technology Group, a China-based company with links to the Chinese government, as the enabler of a hacking campaign against organisations worldwide [2] (p.1). That is a genuinely large, genuinely coordinated piece of attribution work. The NCSC's own announcement of it, though, describes the campaign in terms the advisory it is summarising does not support.
The claim: "uniquely using AI tools"
The NCSC's news page is direct about what makes this campaign stand out. "Malicious cyber actors, enabled by Integrity Tech, are uniquely using AI tools, such as automated scanning, alongside large-scale botnets and manual exploitation techniques to compromise and steal confidential data from companies around the world, including critical sectors," it says [1]. The same page's opening summary repeats the framing: "AI-enabled tools, large-scale botnets and hands-on exploitation techniques being used to compromise networks and steal sensitive data" [1]. "Uniquely" is doing real work in that sentence: it tells a reader this is not an ordinary hacking campaign but something that marks Integrity Tech out, because of AI.
The NCSC's page links the full technical advisory, product ID AA26-281A, as the source for all of this [1]. That 58-page document, co-authored by the NCSC itself, is the evidence the claim rests on.
What the advisory the NCSC co-wrote actually says
The advisory's own executive summary describes the campaign as "combining automated scanning tools, large-scale botnets, and hands-on exploitation techniques" that "exploit vulnerabilities by using scanning tools, cross-site scripting attacks, and password spraying on Microsoft Exchange servers" [2] (p.2). No mention of AI there either. Across all 58 pages, including the technical details, mitigations, MITRE ATT&CK mapping and two appendices of indicators of compromise, the words "AI," "artificial intelligence" and "machine learning" do not appear once.
The reconnaissance section names the actual tools: "BBScan, dirsearch, Fscan, ksubdomain, masscan, Nmap, OneForAll, ShuiZe, and WPScan," adding that "the use of open source tools typically found on GitHub suggests the threat actors tend to look for more vulnerable targets" [2] (p.5). These are free, publicly documented scanning utilities, widely used by penetration testers and hobbyists; the advisory's own assessment is that relying on them points to opportunism, not sophistication.
The centrepiece tool gets its own subsection: "As early as 2017, these threat actors have also used a malicious application known as 'MicroScan.' This Python-based web application contains over 1,300 penetration testing scripts written to scan websites for specific vulnerabilities," targeting "OpenSSL, Oracle WebLogic Server, Rejetto HFS, WordPress, Juniper ScreenOS, Jenkins, and Apache Struts" [2] (p.5). A Python script that runs pre-written vulnerability checks is automation. It is not, on anything the advisory itself says, artificial intelligence.
The advisory's own Appendix B lists the eight CVEs it says the campaign has successfully exploited: CVE-2014-6278, CVE-2015-3306, CVE-2015-5477, CVE-2016-3081, CVE-2019-11510, CVE-2021-22205, CVE-2021-3199 and CVE-2023-22894 [2] (p.2, p.58). The oldest, a Bash command-injection flaw, is twelve years old; the newest, in the Strapi content management system, is three. Six of the eight carry the advisory's own asterisk marking them as entries "newly added to CISA's Known Exploited Vulnerabilities (KEV) Catalog" [2] (p.58), meaning they are flaws that were already known and already had patches available, not novel exploits the campaign discovered. The advisory's mitigation advice matches that picture: "disabling unused services and ports," "sanitizing web application inputs," requiring multifactor authentication, and "applying timely patches" [2] (p.2). Standard hygiene against years-old, patchable vulnerabilities is not a defence against an AI threat; it is a defence against organisations that have not patched.
The claims, tested
| The document's own words | What the record shows | Verdict |
|---|---|---|
| "Malicious cyber actors, enabled by Integrity Tech, are uniquely using AI tools... to compromise and steal confidential data" (NCSC news page, 8 Oct 2026) [1] | The 58-page advisory the NCSC co-authored and cites as its source never uses the words AI, artificial intelligence or machine learning anywhere in its text [2] | Unsupported by the document's own cited evidence |
| The campaign's core tool, described by the NCSC as an "AI tool" | The advisory describes it as "a malicious application known as 'MicroScan'," "a Python-based web application" of "over 1,300 penetration testing scripts," in use "as early as 2017" [2] (p.5) | Scripted automation, not AI, and nine years old |
| Implied novelty and sophistication behind the "uniquely" framing | The advisory's own Appendix B lists eight exploited CVEs dated 2014 to 2023, six already added to CISA's Known Exploited Vulnerabilities catalogue [2] (p.58) | Known, patchable flaws, not novel tradecraft |
Credit where due
The underlying investigative work is substantial and the advisory is transparent about its own limits. Eight agencies across seven countries coordinated a single technical document naming a specific enabling company, mapping its techniques to the MITRE ATT&CK framework, and publishing two full appendices of indicators of compromise so that network defenders can act on them directly [2]. The advisory does not oversell its own indicators either: it warns readers directly that "several of the observed IOCs date back to as early as 2016" and recommends "investigating or vetting these IOCs prior to taking action, such as blocking" [2] (p.18), an honesty about stale data that the NCSC's AI framing does not extend to the advisory's headline claim. The "last year" reference to the UK sanctioning Integrity Tech also checks out: the Foreign, Commonwealth and Development Office's sanctions announcement is dated 9 December 2025 [3], correctly "last year" from an October 2026 announcement.
Verdict
Four stars, from one documented mistake. The underlying advisory is solid, multilateral, technical work; the fault sits entirely in how the NCSC's own announcement characterised it. Calling a nine-year-old Python scanning toolkit and a list of mostly years-old, already-catalogued vulnerabilities "uniquely" AI-powered does not make the threat more serious. It makes the warning easier to dismiss once a reader checks the document the NCSC itself says to read.
The star score counts one documented mistake: the NCSC's news page describes the campaign as "uniquely using AI tools" and "AI-enabled," a characterisation the 58-page joint advisory it co-authored and cites as its evidence never once supports, in text that instead names a Python penetration-testing toolkit in use since 2017 and eight exploited vulnerabilities dated 2014 to 2023. One fall in the 1 to 3 band: four stars; the bands are on the ratings page. This piece makes no finding about the accuracy of the advisory's technical content itself, which this piece did not independently test against victim organisations, nor about the good faith of the NCSC or its seven co-authoring agencies; the mismatch identified here is narrower, between one characterisation on the NCSC's own announcement page and the document that announcement cites as its source. Checked directly against the full text of the NCSC's announcement and the complete 58-page advisory (product ID AA26-281A), both fetched 9 October 2026.
Sources
- China-linked malicious actors called out by UK and international partners for targeting sensitive data globally, National Cyber Security Centre, published 8 October 2026
- Chinese Government-linked Cyber Threat Actors Combine Automated and Hands-on Hacking Tools to Steal Sensitive Data, joint advisory AA26-281A, FBI, CISA, NSA, NCSC-UK, Australian Cyber Security Centre, Canadian Centre for Cyber Security, Japan's National Police Agency and National Cybersecurity Office, New Zealand NCSC and Spain's Centro Nacional de Inteligencia, published 8 October 2026
- UK clamps down on China-based companies for reckless and irresponsible activity in cyberspace, Foreign, Commonwealth and Development Office, published 9 December 2025