UK Security Service warns of growing national security threats from Russia and Iran
Spread of AI-enabled cyberattacks exposes security vulnerabilities in the British government and businesses
Changing attack methods broaden threats, highlighting the limitations of conventional security approaches

Concern over national security threats is growing in the UK. As attempted sabotage and cyberattacks backed by the governments and intelligence agencies of hostile states such as Russia and Iran increase, advances in artificial intelligence (AI) are accelerating cyberattacks and expanding their reach, heightening concerns about the defensive capabilities of the British government and businesses. Experts say minimizing AI-driven threats requires moving beyond security systems centered on conventional firewalls and antivirus software to establish an environment capable of detecting and responding to anomalous behavior in real time.

A warning from the head of the UK Security Service

According to a report by The Guardian on the 13th (local time), Ken McCallum, director general of the UK Security Service (MI5), said in his annual threat assessment speech at MI5 headquarters in London that day that national security threats linked to foreign governments were expanding rapidly. “As of 2026, MI5 is devoting more resources to countering state threats than it did during the Cold War,” McCallum said. “Our caseload involving state threats has quadrupled over the past 10 years.” He added that cases involving Russia, Iran and China had led to prosecutions and convictions, and that dozens of people had been arrested in the three years since the National Security Act took effect in 2023.

McCallum identified Russia as the “number one aggressor” threatening British security. “Putin’s Kremlin continues its aggressive behavior toward the UK and our allies,” he said. “Russia’s military and intelligence agencies are seeking to expand their hybrid warfare while avoiding circumstances that would trigger Article 5, NATO’s collective defense clause.” The threat from Iran was also cited as a major concern. “More than 50 people have been arrested in the UK over the past 12 months in connection with Iranian-linked operations,” McCallum said, noting that Iran was using criminal organizations and others to threaten dissidents, media organizations and Jewish communities. This year in particular, there have reportedly been signs of efforts to move beyond existing threats, arson and sabotage to target heavily guarded sites such as military facilities.

AI-driven cybersecurity threats intensify

McCallum had previously warned about the risks of hostile states using AI. The Times reported on the 8th of last month that McCallum, Secret Intelligence Service (MI6) chief Blaise Metreweli and Government Communications Headquarters (GCHQ) director Anne Keast-Butler attended a Cabinet meeting and discussed the security risks posed by the misuse of AI technology. They warned that not only terrorist groups and individual terrorists, but also hostile states such as Russia and Iran, could use AI to undermine British national security. McCallum reportedly described AI-driven threats as “the biggest issue ministers need to think about.”

The warnings from Britain’s intelligence chiefs came amid growing concerns about the government’s cyber defense capabilities. In June, the Government Cyber Coordination Centre (GC3) announced that it had identified a total of 407 security vulnerabilities after using advanced AI models to analyze publicly available software code from 9 government bodies over one month. These included serious vulnerabilities that could allow external attackers to bypass system authentication or execute malicious code remotely. The AI models used in the review cost just GBP 13,000 (approximately USD 17,769). One British Cabinet official said the government was failing to respond adequately despite repeated warnings, comparing the current situation to the early stages of the COVID-19 pandemic and the 2008 global financial crisis.

Damage mounts at businesses and institutions

Research also shows that AI-enabled cyberattacks have caused actual damage to British businesses and institutions. According to the “2026 Cost of a Data Breach Report” released by US company IBM in July, AI-enabled attacks caused 22% of malicious data breaches at the British businesses and institutions surveyed. In another 13% of incidents, investigators could not determine whether AI had been involved. The study analyzed data breaches at 602 businesses and institutions worldwide between March last year and February this year, including 48 organizations in the UK.

Globally, AI was used in 1 in 4 malicious data breaches, while the number of AI-based attacks rose 56% year over year. In addition, 62% of AI-enabled attacks targeted critical infrastructure, with particularly high concentrations in finance and energy. Attack methods have also diversified. IBM found that deepfake attacks, which impersonate people by precisely mimicking their faces or voices, accounted for the largest share of AI-enabled attacks at 45%. These were followed by attacks that used AI to generate or modify malware at 19%, and phishing attacks that used AI to craft persuasive messages to steal victims’ account details or authentication credentials at 17%.

Table 1. Global landscape of AI-based cyberattacks

US authorities also monitoring the risks

As AI-driven security threats become more apparent, major countries beyond the UK are increasingly taking notice of the risks. For example, the US Federal Trade Commission (FTC) recently sent civil investigative demands (CIDs) to OpenAI and Anthropic. A civil investigative demand is an administrative procedure comparable to a court subpoena, issued when authorities seek internal corporate records and other materials. In demands totaling 50 pages, the FTC reportedly requested materials concerning recent security incidents and related safeguards, focusing on potential consumer protection violations. It also plans to summon executives from both companies soon to hear testimony on how AI products could affect US consumers.

The move is seen as a response to recent instances of aggressive, anomalous behavior by OpenAI and Anthropic models. A prominent example occurred in July, when OpenAI’s so-called “rogue agents” left their isolated environment without authorization and attacked the systems of an external organization, Hugging Face. OpenAI CEO Sam Altman described it as the most serious instance of anomalous behavior identified in the company’s AI models to date. More recently, it emerged that an OpenAI agent had leaked 53 images belonging to an individual ChatGPT user online and hacked an Australian government health statistics website. It also belatedly became known that an Anthropic AI model had generated a false tip concerning a murder and submitted it in July to “Philly Unsolved Murders,” a website operated by the Philadelphia Police Department for tips on unsolved homicides.

The burden on attackers is rapidly shrinking

Governments are stepping up their vigilance because advances in AI technology have lowered the technical and financial barriers to launching cyberattacks. According to a report released last month by the Google Threat Intelligence Group (GTIG), cyberattackers are increasingly using AI agents to automate vulnerability searches, the execution of attack tools and credential collection. These changes have not only made cyberattacks substantially easier to carry out but have also affected their economics. In a recently published threat intelligence report, Anthropic said that greater AI autonomy reduces the technical expertise, labor and costs required for cyberattacks. As a result, even small organizations previously excluded as targets because expected returns were too low relative to the cost of an attack could become new targets.

As cyberattacks become faster and broader in scope, the limitations of conventional security approaches are also becoming clearer. Traditional firewalls and antivirus programs are effective at detecting abnormal external access or malware, but they cannot easily block every attack that exploits legitimate user accounts, cloud services or authorized administration tools. In particular, when AI rapidly identifies vulnerabilities and automates attacks, security personnel have less time to spot warning signs and respond. Security industry experts therefore say organizations need to go beyond protecting individual systems and build security frameworks that integrate the management of user devices, servers, cloud environments and account authentication systems, while detecting anomalous behavior and enabling real-time incident response.