SafetyRisk, alignment & guardrails
NCSC says AI-powered malware is 'highly likely' to be used in a ransomware attack on UK critical infrastructure by 2028
The UK's National Cyber Security Centre judges it highly likely that AI-powered malware will feature in a ransomware attack on critical national infrastructure by 2028. Its view is that AI helps attackers exploit long-tolerated legacy weaknesses at scale rather than inventing new attacks.
From the The Guardian · AI article of November 3, 2026 and cited sources: The NCSC Annual Review 2026 forecasts that AI-powered malware will highly likely be used in a ransomware attack on UK critical national infrastructure, such as energy, health and transport, by 2028. [1] [10]
SOURCES[1] AI-powered malware ‘highly likely’ to be used to hack UK… theguardian.com[10] Infosecurity Magazine: NCSC Chief Warns of AI Ransomware… infosecurity-magazine.com
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- The NCSC Annual Review 2026 forecasts that AI-powered malware will highly likely be used in a ransomware attack on UK critical national infrastructure, such as energy, health and transport, by 2028. [1] [10]
- Original source
- The Guardian · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 3, 2026 article published
- Covered
- Nov 3 edition
- Evidence
- 10 sources across 8 sites
- 1AI-powered malware ‘highly likely’ to be used to hack UK…theguardian.com
- 2Chinese and Russian AI cyber attacks will be routine by…ft.com
- 3Blockonomi: UK Cyber Agency Warns AI Ransomware Attack on…blockonomi.com
- 4NCSC: Richard Horne's speech at the Annual Review 2026…ncsc.gov.uk
- 5NCSC: Impact of AI on cyber threat from now to 2027ncsc.gov.uk
- 6NCSC: UK experiencing four 'nationally significant' cyber…ncsc.gov.uk
Reach: not measured yet
The NCSC Annual Review 2026 forecasts that AI-powered malware will highly likely be used in a ransomware attack on UK critical national infrastructure, such as energy, health and transport, by 2028. It also expects Russia and China to use AI routinely in cyber operations. [1] [10]
NCSC chief Richard Horne said AI's main effect is to let attackers find and exploit old, tolerated weaknesses at scale, and blamed decades of cheaply built technology. This follows the agency's 2025 assessment that AI mostly strengthens existing techniques and that fully automated end-to-end attacks were unlikely before 2027. [4] [5]
In the php2shell case, AI helped find two use-after-free flaws that affect PHP releases back to version 5.4. The credit for AI's role comes from Wiz and OpenAI themselves. After disclosure, some Finnish and Dutch government sites went offline while they patched, and the NCSC sent more than 110,000 notifications and early-warning alerts. [8] [9] [4]
The judgement is an intelligence forecast, not an observed attack. Still, the AI-assisted php2shell disclosure showed how quickly old memory-safety flaws in widely used software can force emergency patching.
Read the full assessment
That suggests operators of legacy systems will have less time to fix flaws.
I'm writing the dossier now from the NCSC speech, the 2027 AI assessment, Wiz's technical write-up and the other reporting.
Executive brief
The UK's cyber agency now rates it "highly likely" that AI-powered malware will be used in a ransomware attack on UK critical national infrastructure (CNI) by 2028. CNI covers sectors such as energy, health and transport. The warning comes from the NCSC Annual Review 2026. On UK intelligence's probability scale, "highly likely" is roughly an 80–90% chance, according to Blockonomi. NCSC chief Richard Horne warned of a possible "unprecedented rise" in incidents within months. His reasoning is that AI lets attackers find and exploit old, tolerated weaknesses at scale, not that it invents new kinds of attack (NCSC speech).
What changed and event timeline
NCSC sets its AI baseline
The NCSC judged AI would highly likely make intrusions more frequent and more damaging by 2027. It called fully automated end-to-end advanced attacks unlikely before 2027 ().
Significant incidents more than double
The NCSC handled 204 "nationally significant" incidents in the year to August 2025, up from 89 the year before ().
AI finds PHP flaws
Calif, OpenAI and Wiz found the memory-safety bugs later named "php2shell" with AI help. Wiz says it reported them to the PHP security team on 24 August (;).
PHP patches; sites go offline
Patched releases and public disclosure followed. Finland and the Netherlands took some government websites offline while they applied fixes ().
Annual Review 2026
The review makes the 2028 CNI ransomware judgement and says Russia and China will use AI "routinely." Horne's launch speech blames decades of "tech on the cheap" ().
Capabilities and access
- The PHP discovery used three named systems: Wiz Atlas, Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol (Wiz).
- The flaws are CVE-2026-109618 and CVE-2026-109619. Every PHP release back to version 5.4 is affected by at least one of them (Wiz).
- The Guardian reports that OpenAI and Anthropic have given some companies early access to their strongest models for finding vulnerabilities.
Read the full section
- The PHP discovery used three named systems: Wiz Atlas, Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol (Wiz).
- The flaws are CVE-2026-109618 and CVE-2026-109619. Every PHP release back to version 5.4 is affected by at least one of them (Wiz).
- The Guardian reports that OpenAI and Anthropic have given some companies early access to their strongest models for finding vulnerabilities. It does not name the programmes or say who qualifies.
Technical analysis for researchers and developers
- What AI did in the PHP case. Humans adapted the exploits to different PHP versions, configurations and memory layouts.
- Both flaws are use-after-free bugs, in the upload-progress unserializer and in fatal-error backtrace handling. This is the class of legacy memory-safety debt Horne highlights.
- Horne says the Linux kernel project patched more than 1,000 vulnerabilities in a single month (NCSC speech).
Read the full section
- What AI did in the PHP case. Wiz credits AI with help across four stages: finding the bug, developing the exploit, detecting attacks and checking the patch. Humans adapted the exploits to different PHP versions, configurations and memory layouts. Wiz says LLMs cut the specialist effort needed for each target (Wiz).
- The bug type. Both flaws are use-after-free bugs, in the upload-progress unserializer and in fatal-error backtrace handling. This is the class of legacy memory-safety debt Horne highlights.
- Patch volume. Horne says the Linux kernel project patched more than 1,000 vulnerabilities in a single month (NCSC speech). The public summaries do not include an NCSC method for measuring how much of that discovery is AI-driven.
Claims and evidence
- The 2028 CNI ransomware judgement. This is an intelligence forecast, not an observed event (Guardian).
- No fully autonomous attack so far. This appears only in secondary summaries (Blockonomi).
- Horne says the NCSC sent more than 110,000 notifications and early-warning alerts over php2shell (NCSC speech).
Read the full section
- The 2028 CNI ransomware judgement. This is an intelligence forecast, not an observed event (Guardian).
- No fully autonomous attack so far. The review reportedly says AI acts as a force multiplier and that no fully autonomous attack has been recorded. This appears only in secondary summaries (Blockonomi).
- NCSC alert volume. Horne says the NCSC sent more than 110,000 notifications and early-warning alerts over php2shell (NCSC speech).
- Who found the flaws. The model credits come from Wiz and OpenAI themselves. Nobody independent has evaluated how much AI contributed.
Context and prior work
- Consistent with 2025. The 2028 judgement follows the NCSC's 2025 view that AI mainly strengthens existing attack techniques rather than creating new ones (NCSC assessment).
- The 2025 assessment warned of a gap between systems that keep pace with AI-enabled threats and a large share that become more exposed.
- Earlier UK ransomware victims include Royal Mail, Capita and several councils (Guardian).
Read the full section
- Consistent with 2025. The 2028 judgement follows the NCSC's 2025 view that AI mainly strengthens existing attack techniques rather than creating new ones (NCSC assessment).
- The "digital divide." The 2025 assessment warned of a gap between systems that keep pace with AI-enabled threats and a large share that become more exposed. It rated universal access to AI for defence by 2027 a "remote chance."
- Earlier UK ransomware victims include Royal Mail, Capita and several councils (Guardian).
Limitations, safety and contested findings
- Timing is disputed. Bridewell CEO Anthony Young said the two-year horizon "feels generous" because AI is already helping attackers (Infosecurity).
- OpenAI said the capabilities that found php2shell could also help attackers build exploits quickly (Fortune).
- Surrey University's Alan Woodward warned that if defenders cannot get these tools, attackers will find the flaws first (Guardian).
Read the full section
- Timing is disputed. Bridewell CEO Anthony Young said the two-year horizon "feels generous" because AI is already helping attackers (Infosecurity). The NCSC's own wording says AI mainly improves existing operations rather than running attacks end to end.
- The same tools cut both ways. OpenAI said the capabilities that found php2shell could also help attackers build exploits quickly (Fortune).
- Uneven access to defensive AI. Surrey University's Alan Woodward warned that if defenders cannot get these tools, attackers will find the flaws first (Guardian).
Business and practitioner implications
- What the NCSC asks organisations to do: replace end-of-life technology, adopt passkeys, secure physical control systems, and fund monitoring and detection (NCSC speech).
- New government funding includes £60m for Cyber Shield phase one, £120m more for the AI Security Institute and £40m for a National Agentic Security Unit.
- Patching has to get faster. php2shell showed a widely used platform can need emergency fixes everywhere at once.
Read the full section
- What the NCSC asks organisations to do: replace end-of-life technology, adopt passkeys, secure physical control systems, and fund monitoring and detection (NCSC speech).
- New government funding includes £60m for Cyber Shield phase one, £120m more for the AI Security Institute and £40m for a National Agentic Security Unit. Cyber Essentials certification gets a 75% discount (Infosecurity).
- Patching has to get faster. php2shell showed a widely used platform can need emergency fixes everywhere at once.
The source trail.
Sources (10)
AI-powered malware ‘highly likely’ to be used to hack UK infrastructure by 2028
Article text retrieved; extracted text may omit tables or interactive elements.
theguardian.comChinese and Russian AI cyber attacks will be routine by 2028, says GCHQ
Related coverage; assess separately
ft.com