The UK's National Cyber Security Center is warning that AI-powered cyberattacks on critical infrastructure are highly likely by 2028. Head of the NCSC, Richard Horne, says criminals and state actors like China and Russia are already leveraging AI to enhance malware, scout targets, and exploit network weaknesses. The annual review highlights that AI will more likely sharpen existing attack methods than invent entirely new vulnerabilities, and that Russia is conducting persistent cyber campaigns below the threshold of armed conflict.
The financial toll is already significant. Malicious cyber activity costs UK organizations nearly $20 billion per year, and automaker Jaguar Land Rover suffered over $260 million in direct losses from a 2025 cyberattack. A December 2025 attack on Poland's energy grid, attributed to Russia's FSB, underscores how real these threats have become. GCHQ director Anne Keast-Butler describes cybersecurity as part of the nation's front line defense.
To counter these threats, the NCSC is developing defenses including Project DROMOS and Cyber Shield, an agentic defense system to be delivered over five years. Funding includes £40 million for a National Agentic Security Unit and £24 million for public sector defenses. Open source AI security tools like ARTEX, a Chinese-made red-teaming platform, are already being repurposed by attackers, and China is viewed as the most sophisticated and sustained cyber challenge facing the UK.
On the enterprise side, Heimstaden, a European residential real estate operator, turned to OpenAI tools to modernize property management. Using ChatGPT and Codex, the company cut property image production costs by 99 percent and analyzed 20,112 property assets. Employee adoption reached 50 percent within two to three months, showing that AI tools can deliver rapid results when workers are involved in reviewing and approving decisions.
Walmart is also expanding its AI capabilities. Its agent called Sparky now lets users upload pictures or screenshots to generate shopping lists, find items in stores, and get recipe suggestions. Sparky can speak in Spanish and French, and Walmart plans to add voice capabilities by year-end while expanding availability to Canada and Mexico.
Safety concerns persist as well. A recent study found that predictive safety guardrails do not always make large language model agents safer. Researchers tested seven policy models and found that stronger prediction involvement did not consistently improve safety, and world-model predictions become less reliable over longer time periods. Better predictions alone do not guarantee safer decisions, raising questions about how to secure AI systems as they take on more autonomous roles.
Key Takeaways
- The UK's NCSC warns that AI-powered cyberattacks on critical infrastructure are highly likely by 2028, with China and Russia as primary threats
- Malicious cyber activity costs UK organizations nearly $20 billion annually, and Jaguar Land Rover lost over $260 million in a 2025 attack
- The NCSC is developing Cyber Shield, an agentic defense system, backed by £40 million for a National Agentic Security Unit and £24 million for public sector defenses
- Heimstaden used OpenAI's ChatGPT and Codex to cut property image production costs by 99 percent and analyze over 20,000 property assets
- Walmart's Sparky AI agent now converts pictures into shopping lists and supports Spanish and French, with voice capabilities coming soon
- A study found that AI safety guardrails do not always improve safety in large language model agents, especially over longer time periods
- Russia is conducting cyber campaigns below the threshold of armed conflict, while China is considered the most sophisticated cyber threat to the UK
- Open source AI tools like ARTEX, a Chinese-made red-teaming platform, are being repurposed by attackers
- Jodie Cook's AI clone managed 5,000 coaching conversations in one month, highlighting the trend of scaling personal expertise through AI
- China's first fully AI-generated film, Sanxingdui: Future Past, earned just 1.7m yuan despite costing tens of millions to produce
UK cyber chief warns of AI cyberattacks on infrastructure by 2028
The UK's National Cyber Security Center warns that AI cyberattacks on critical infrastructure are likely by 2028. Richard Horne, head of the NCSC, said criminals and states like China and Russia are increasingly using AI in cyber activity. He warned that AI-powered malware could exploit long-tolerated vulnerabilities at an unprecedented scale. The warning came during the NCSC's annual cyber report release.
Britain warned as AI ransomware attacks expected by 2028
The UK's National Cyber Security Centre says AI-powered malware is highly likely to be part of a future ransomware attack on critical infrastructure by 2028. Its annual review warns that open source AI security tools like ARTEX, a Chinese-made red-teaming platform, are already being repurposed by attackers. China is seen as a key supplier of tools that help attackers hide their activities. The NCSC is also developing defenses including Project DROMOS and Cyber Shield, an agentic defense system to be delivered over five years. Funding includes 40 million pounds for a National Agentic Security Unit and 24 million pounds for public sector defenses. NCSC CEO Dr. Richard Horne urged organizations to invest with urgency in cybersecurity.
UK warned of AI cyberattacks on critical infrastructure by 2028
The UK's National Cyber Security Center warned that AI cyberattacks on critical infrastructure will likely occur by 2028. The report says AI will more likely enhance existing attack methods rather than create brand new vulnerabilities. A December 2025 attack on Poland's energy grid attributed to Russia's FSB security service highlights the threat. Malicious cyber activity costs UK organizations nearly 20 billion dollars per year. Carmaker Jaguar Land Rover suffered over 260 million dollars in direct losses from a 2025 cyberattack.
UK cyber chief says AI cyberattacks on infrastructure coming by 2028
Britain's cyber chief Richard Horne says the UK will likely face AI cyberattacks on critical infrastructure by 2028. He warned that leaders who have gotten away with tech on the cheap will need to explain their failures. Chinese cyber activity is called the most sophisticated and sustained challenge facing the UK. Russia is conducting cyber campaigns below the threshold of armed conflict, while Iran targets individuals threatening the Iranian regime. The UK faces nearly 20 billion dollars in annual cybercrime costs, and Jaguar Land Rover lost over 260 million dollars in a 2025 attack. GCHQ director Anne Keast-Butler said cybersecurity is now part of the nation's front line defense.
AI-powered malware highly likely to hit UK infrastructure by 2028
The National Cyber Security Centre expects AI-powered malware to be used in ransomware attacks against UK power stations, airports, and hospitals by 2028. The NCSC, part of the GCHQ spy agency, warned that Russia and China will use AI routinely in cyber-attacks. AI is expected to help attackers scout targets, find network weaknesses, and write malicious code. NCSC chief Richard Horne will warn that AI will unleash a wave of cyber-attacks imminently. Communications networks used by the British military face daily persistent attacks, with Russia intentionally carrying out campaigns that fall below the threshold of armed conflict.
How CISOs can secure agentic AI environments without slowing innovation
Enterprises face new security risks as AI agents gain autonomy and access across company and personal devices. CISOs need better visibility, clear limits on agent access, and accountability for each agent. Huawei showcased its Security Operations Centre solution at GovWare 2026, showing how Agentic AI can help detect threats. Key steps include inventorying all agents, restricting data access, and using tools like Model Context Protocol servers.
Heimstaden uses OpenAI tools to transform property management
Heimstaden, a European residential real estate operator, chose OpenAI in June 2025 to modernize how it manages thousands of homes. The company uses ChatGPT and Codex to handle pricing, advertising, and listing reviews, cutting property image production costs by 99 percent. Employees review and approve all AI-assisted decisions. Heimstaden analyzed 20,112 property assets with Codex and reached 50 percent employee adoption within two to three months.
Sign of the Times: What is an (AI) author?
This article examines what it means to call an AI system an author, drawing on philosopher Roland Barthes and ideas about plagiarism. AI writing tools raise concerns about originality because they mix probable word patterns from training data rather than producing truly new content. The piece argues that AI undermines the concept of authorship without making readers stronger or more engaged. The author calls for insisting on reading and writing for ourselves.
China's first AI-generated film flops but is seen as cinema's future
Sanxingdui: Future Past became China's first fully AI-generated film, earning just 1.7m yuan at the box office despite costing tens of millions to make. The film used an in-house AI scriptwriting tool called Bowen and AI-created visuals, while humans handled editing. Audiences found it confusing and less entertaining than expected. Still, industry workers believe AI films will keep being made because investors see them as cheaper and the future of cinema.
Jodie Cook Built an AI Clone That Managed 5,000 Coaching Conversations While She Slept
Jodie Cook created an AI version of herself called Jodie AI, trained on her work and ideas, to handle coaching questions she did not have time to answer. The AI clone managed 5,000 coaching conversations in a single month. Cook built it to share her existing books, courses and articles in a more interesting way. The project highlights a shift toward scaling personal access to how someone thinks.
Bradbury Science Museum hosts free AI night event
The Bradbury Science Museum will host 'Night With A Nerd: Neuromorphic Artificial Intelligence' on Wednesday, Nov. 4, from 6 to 8 p.m. Dr. Jeff B. Phillips from Los Alamos National Laboratory will give the presentation. The event is free and open to the public. People can call (505) 665-3382 for more information.
Walmart's Sparky AI now turns pictures into shopping lists
Walmart updated its AI agent called Sparky with new features for the holiday shopping season. Users can upload a picture or screenshot to ask Sparky questions, or turn a handwritten list into a shopping cart. Sparky can also help find items in stores, generate recipes, suggest healthier alternatives, and speak in Spanish and French. Walmart introduced Sparky in June 2025 and plans to add voice capabilities by the end of the year, with availability expanding to Canada and Mexico.
ESUT partners with Erudite Academy on AI training
Enugu State University of Science and Technology signed an agreement with Erudite Future Skills and Technology Academy to train students and staff in artificial intelligence, cybersecurity, robotics and 3D technologies. The partnership includes student internships, staff professional development and joint research. Prof. Gabriel Zeruwa Ugwu and Prof. Michael Edeh helped coordinate the signing ceremony. The collaboration aims to connect university learning with industry needs and prepare participants for jobs in the digital economy.
Study finds AI safety guardrails do not always work
A new study found that predictive safety guardrails do not always make large language model agents safer. Researchers tested seven policy models across two safety environments and found that stronger prediction involvement did not always lead to better safety. The study showed that world-model predictions become less reliable over longer time periods. Safety gains can still happen even with corrupted predictions, showing that better predictions alone do not guarantee safer decisions.
Sources
- UK cyber chief says the UK will face AI cyberattacks on critical infrastructure by 2028
- Britain on notice: AI-powered ransomware attacks to arrive by 2028
- UK warned of AI cyberattacks on critical infrastructure by 2028
- UK cyber chief says UK will face AI cyberattacks on critical infrastructure by 2028
- AI-powered malware ‘highly likely’ to be used to hack UK infrastructure by 2028
- How CISOs can secure agentic AI environments without slowing innovation
- Heimstaden is changing residential real estate management with AI
- Sign of the Times: What is an (AI) author?
- China’s first AI-generated film flops but still seen as the future of cinema
- Jodie Cook Built an AI Clone That Managed 5,000 Coaching Conversations While She Slept
- Bradbury Science Museum To Present ‘Night With A Nerd: Neuromorphic Artificial Intelligence’ 6-8 p.m. Wednesday Nov. 4
- Walmart’s Updated AI Agent Turns Pictures Into Shopping Lists
- ESUT, Erudite Academy partner on AI, cybersecurity training
- Predictive Safety Guardrails Do Not Always Make LLM Agents Safer
Comments
Please log in to post a comment.