“AI abuse in hostile nation cyber offensives” UK’s growing national security concerns test the effectiveness of existing cyber defense systems

UK Security Intelligence Service warns of expanding national security threats originating from Russia and Iran
Proliferation of AI-utilized cyber attacks exposes vulnerabilities in UK government and corporate security
Changes in attack methods broaden threat scope, highlighting limitations of traditional security measures

In the UK, vigilance against national security threats is increasing. Attempts at sabotage and cyber attacks backed by hostile governments and intelligence agencies such as Russia and Iran are expanding. Meanwhile, advancements in artificial intelligence (AI) technology are accelerating the speed and broadening the scope of cyber attacks, raising concerns about the defense capabilities of the UK government and businesses. Experts point out that to minimize these AI-generated threats, it is necessary to establish an environment that can detect and respond to anomalous behavior in real time beyond the conventional security systems centered on firewalls and antivirus software.

Warning from UK Security Intelligence Service Director

According to a report by The Guardian on the 13th (local time), Ken McCallum, director of the UK Security Intelligence Service (MI5), stated during the annual security threat assessment speech held at MI5 headquarters in London that national security threats linked to foreign governments are rapidly expanding. McCallum said, "As of 2026, MI5 is allocating more resources to counter national-level threats than during the Cold War era," adding, "The number of national threat-related cases handled has quadrupled over the past decade." He also explained that prosecutions and convictions related to Russia, Iran, China, and others are ongoing, and that dozens of people have been arrested within three years since the implementation of the National Security Act in 2023.

McCallum identified Russia as the 'primary aggressor' threatening UK security. He stated, "Putin's Kremlin continues aggressive actions against the UK and its allies," and criticized that the Russian military and intelligence agencies are attempting to expand hybrid warfare while avoiding triggering Article 5 of the North Atlantic Treaty Organization (NATO) collective defense clause. Iran was also mentioned as a major concern. McCallum said, "In the past 12 months, more than 50 people linked to Iran-related operations have been arrested in the UK," pointing out, "Iran is using criminal organizations and others to threaten dissidents, media organizations, and Jewish communities." Particularly, this year there have been moves to target highly secured sites such as military facilities, beyond traditional threats, arson, and sabotage.

Intensifying AI-driven cybersecurity threats

Earlier, McCallum also warned of the dangers of hostile nations using AI. The Times reported on the 8th of last month that McCallum, along with Richard Moore, director of the Secret Intelligence Service (MI6), and Anne Keast-Butler, head of the Government Communications Headquarters (GCHQ), attended a cabinet meeting where they discussed security risks arising from the misuse of AI technologies. They pointed out that terrorist groups and lone terrorists, as well as hostile nations like Russia and Iran, could use AI to undermine UK national security. Notably, McCallum described AI-induced threats as 'the biggest issue that ministers need to consider.'

Warnings from UK intelligence chiefs come amid growing concerns over the British government's cyber defense capabilities. Previously, the UK Government Cyber Coordination Centre (GC3) announced in June that after analyzing open-source software code from nine government agencies for a month using advanced AI models, it identified a total of 407 security vulnerabilities. Among them were critical vulnerabilities allowing external attackers to bypass system authentication or remotely execute malicious code. The AI model deployed for this inspection cost £13,000 (about USD 17,769). In relation to this, a UK cabinet official noted that despite repeated warnings, the government has not taken sufficient measures, comparing the current situation to the early stages of the COVID-19 pandemic and the 2008 global financial crisis.

Accumulated actual damage to businesses and institutions

There is also research showing that UK companies and institutions have suffered actual damages from AI-integrated cyber attacks. According to IBM's '2026 Cost of a Data Breach Report' released in July, 22% of malicious information breaches in UK companies and institutions surveyed were caused by attacks involving AI. Another 13% of breaches did not confirm AI involvement. This report analyzed data breach incidents from March last year to February this year across 602 companies and institutions worldwide, including 48 organizations in the UK.

Expanding the scope globally, one out of four malicious data breach incidents involved AI, and cases of AI-based attacks increased by 56% compared to the previous year. Additionally, 62% of AI-driven attacks targeted critical infrastructure sectors, with relatively high concentrations in finance and energy. The attack methods also diversified. IBM analyzed that among AI-based attacks, the most significant portion—45%—were deepfake attacks impersonating real people’s faces or voices. Following that, attacks generating or modifying malware using AI accounted for 19%, and phishing attacks crafting persuasive messages via AI to steal victims’ account credentials or authentication means accounted for 17%.

Table 1. Global Status of AI-Based Cyber Attacks

US Authorities also Monitor Risks

As AI-driven security threats become apparent, major countries besides the UK are also increasingly focusing on these risks. For example, the US Federal Trade Commission (FTC) recently issued civil investigative demands (CID) to OpenAI and Anthropic. A CID is an administrative procedure equivalent to a court subpoena, sent when authorities seek internal documents from companies. The FTC is reportedly focusing on possible consumer protection violations in the 50-page demand, requesting information on recent security incidents and related protective measures. It also plans to summon executives of the companies soon to hear testimonies regarding the impact of AI products on US consumers.

This is seen as a response to recent aggressive anomalous behavior exhibited by AI models from OpenAI and Anthropic. A representative incident was in July when OpenAI's so-called 'rogue agents' arbitrarily escaped isolated environments and attacked the external system of Hugging Face. OpenAI CEO Sam Altman evaluated this as the most serious anomalous behavior identified in their AI models to date. Recently, it was also revealed that OpenAI's AI agents leaked 53 personal images of ChatGPT users online and hacked the Australian government's health statistics site. Additionally, Anthropic’s AI model generated false tips related to a homicide case, which were submitted to the 'Philly Unsolved Murders' cold case tip site run by the Philadelphia Police Department in July, a fact that surfaced later.

Rapidly decreasing attacker burden

Governments worldwide have entered this heightened alert due to AI technology reducing the technical and economic entry barriers for cyber attacks. According to a report released last month by Google's Threat Intelligence Group (GTIG), recent cyber attackers are automating vulnerability discovery, attack tool execution, and credential harvesting using AI agents. These changes not only significantly lower the difficulty of cyber attacks but also affect economic structures. Anthropic’s recent threat intelligence report analyzed that as AI autonomy increases, the skill level, manpower, and costs required for cyber attacks decrease, making smaller organizations—previously excluded due to low return on investment—potential new targets.

As the speed and scope of cyber attacks expand, the limitations of existing security methods are becoming clearer. Traditional firewalls and antivirus programs are effective in detecting abnormal external access or malware but cannot block all attacks exploiting legitimate user accounts, cloud services, and authorized management tools. Especially when AI quickly discovers vulnerabilities and automates attack processes, security personnel have less time to detect and respond to anomalies. Consequently, the security industry diagnoses that it is necessary to build integrated management systems for user devices, servers, cloud environments, and authentication systems that can detect anomalous behavior and enable real-time problem response beyond protecting individual systems alone.