Observe the evidence
Prioritise confirmed essential-service incidents, credible cyber campaigns, actively exploited vulnerabilities and authoritative advisories.
THE THINKING BEHIND THE SIGNAL
Our sources.
Our thinking.
Two separate lenses.
Cyber Threat and AI capability.
Cyber Threat describes evidenced disruption and escalation. The AI Capability Frontier describes breadth and reliability of ability. Neither is an official classification, a probability of a power cut, or a countdown. Higher AI capability does not automatically raise the cyber threat level.
Put it in context
Assess impact, reach, relevance and recency. Deduplicate reporting. Keep source confidence separate from threat severity.
Explain every movement
Publish the factors behind a change, when sources were checked and what is uncertain. Missing data must never silently become an “all clear”.
HOW THE MONITOR WORKS
Feeds and assessments have different roles.
We set both current levels manually as editorial assessments, with their dates and notes available in the scale information panels. Cyber Threat covers the global picture; it is not an automated feed score. Service feeds update independently: an outage, a vulnerability warning and a confirmed cyber incident are different forms of evidence.
Cyber assessments must name their geographic scope. They can move up or down and can skip stages when the evidence warrants it. Missing data means “assessment unavailable”.
CYBER THREAT
A SEPARATE CAPABILITY FRAMEWORK
Understanding the AI frontier.
Our proposed six stages describe capability milestones, including recursive self-improvement (RSI). They do not measure consciousness, intention or safety. A model may excel in one task while remaining limited elsewhere.
“Broad Capacities” sits immediately below RSI, followed by AGI and Superintelligence. RSI describes improvement of capabilities and of the process for further improvement. This sequence is our editorial framework; it is not an established scientific progression. Generality and performance still need separate evidence.
Read the research on levels of AGI ↗OUR FORECAST HAS A DIFFERENT ROLE
Evidence, interpretation and our outlook.
Our five-year outlook is that nearly all paid jobs disappear by October 2031, without a broad return of employment. We use that premise in Life after AGI. It is Blackout Day’s editorial forecast. The sources below explore rapid AI progress, job displacement and cyber risk; they do not establish our exact timeline or job-loss estimate.
That outlook does not automatically raise either live level. The dated Forecast page is a separate scenario. Provider outages describe service availability. Cyber reports describe sourced incidents. Keeping these roles clear lets you see what is reported, what is measured and what we expect.
FOLLOW THE ORIGINAL EVIDENCE
Read the research.
Start with a topic. Each link goes to the original publisher, with a short note on what it tells us and where its limits lie. Reading list reviewed 26 October 2026. Live-feed freshness is shown separately in the monitor’s source details.
01 / 6 SOURCESLive monitoring feeds
These are the connected sources behind the monitor and Disruption Watch. Service status, news and advisories have different roles.
Connected feed · Provider status
Cloudflare System Status
Cloudflare
Monitored network locations inform Country Status. This is provider coverage, not a measure of every service or the national power grid.
Connected feed · Provider status
Google Cloud Service Health
Google Cloud
Active incidents inform service disruption. A cloud incident does not by itself establish a cyber attack.
Connected feed · Official updates
NCSC news and RSS feeds
UK National Cyber Security Centre
Official news and advisories inform Global Watch. A widespread vulnerability warning is not an incident in every country.
Connected feed · Monthly briefings
CERT-EU threat intelligence
CERT-EU
Monthly open-source briefings supply dated reports. We keep incident dates separate from the date a roundup was published.
Connected feed · Original journalism
The Record
Recorded Future News
Headlines help identify developing incidents. A published report is not independent confirmation by Blackout Day.
Connected feed · News discovery
The GDELT Project
GDELT
Helps discover linked coverage. Publisher location and attacker origin are not evidence of where a victim was affected.
02 / 3 SOURCESCyber risk and vulnerability research
These references help explain threat and vulnerability. They do not automatically set our Cyber Threat Level.
Official assessment · AI and cyber risk
Impact of AI on cyber threat from now to 2027
UK National Cyber Security Centre
Assesses how AI may strengthen attackers and widen gaps in defence. It does not forecast a date for a nationwide blackout.
Reference · Observed exploitation
Known Exploited Vulnerabilities Catalog
CISA
Lists vulnerabilities with evidence of exploitation. A listing does not show that every exposed organisation has been compromised.
Reference · Vulnerability model
Exploit Prediction Scoring System
FIRST
Estimates the chance of a published vulnerability being exploited in the next 30 days. It does not predict a household outage.
03 / 3 SOURCESAI capability and safety
Research on what systems can do, how to measure them and what can go wrong. These are reading sources, not automatic inputs to our AI level.
Research paper · 2024
Levels of AGI for Operationalizing Progress on the Path to AGI
Morris and colleagues · ICML 2024
Separates breadth, performance and autonomy. Our six stage names and the position of RSI are our own editorial framework.
Research paper · 2025; revised 2026
Measuring AI Ability to Complete Long Software Tasks
METR and co-authors
Measures task success against the time tasks take people. Selected task results do not measure all jobs or prove a system is generally reliable.
Research review · 2026
International AI Safety Report 2026
International AI Safety Report
Reviews capabilities, misuse, failures and wider risks. It describes uncertainty and differing views rather than one agreed route to AGI.
04 / 4 SOURCESAGI and superintelligence forecasts
Original arguments for rapid change, alongside government planning scenarios. Their dates and assumptions differ.
Forecast scenario · Published 2025
AI 2027
AI Futures Project
Explores AI improving AI and rapid progress to superintelligence. The authors link to updated forecasts; the title is not a fixed deadline.
Forecast essays · June 2024
Situational Awareness: The Decade Ahead
Leopold Aschenbrenner
Argues for a fast path from AGI to superintelligence. It is an individual forecast with substantial assumptions, not an agreed scientific timetable.
Scenario research · AI and physical production
The Industrial Explosion
Tom Davidson and Rose Hadshar · Forethought
Explores AI-directed production and eventual physical automation. The pace depends on robotics, deployment and other assumptions.
Planning scenarios · 2026
AI Scenarios 2030
UK Government
Contrasting futures include rapid capability gains and severe job displacement. These test policy choices; they are not government predictions.
05 / 6 SOURCESJobs, ownership and investment risk
Evidence and models on who earns, who owns and who bears the risk as AI spreads. These sources do not prescribe a household portfolio.
Exposure study · May 2025
Generative AI and jobs: A 2025 update
International Labour Organization
Estimates current generative-AI exposure across job tasks. Exposure is different from job loss, and this is not a study of fully realised superintelligence.
Economic model · 2026
Scenarios for our Economic Future
Anthropic Economics
Explores growth, wages and unemployment under different AI assumptions. Model outcomes are conditional; the publisher also develops AI systems.
Working paper · April 2025
AI Adoption and Inequality
Rockall, Tavares and Pizzinelli · IMF
Models distribution of gains and capital-tax-funded basic income. A modelled payment is not a government commitment to introduce UBI.
Official assessment · July 2026
Financial Stability Report: July 2026
Bank of England
Examines AI investment concentration, infrastructure funding and financial risks. It is not a recommendation to buy AI or data-centre companies.
Investor guidance · Spreading risk
Diversification
Financial Conduct Authority
Explains how spreading investments can reduce reliance on one company or market. Diversification does not remove the risk of losses.
Investor guidance · Land and planning claims
Land banking investment scams
Financial Conduct Authority
Explains risks around land sold on development promises. Owning land does not guarantee planning permission, income or a future profit.
06 / 6 SOURCESHousehold resilience, learning and rights
Practical guidance behind our household actions. Preparation should fit your needs, budget and circumstances.
Household guidance · UK preparedness
Be ready for emergencies
UK Government · Be Ready
Informs water storage and household essentials. The planner uses a drinking allowance; cooking, washing, babies and pets need separate consideration.
Household guidance · Emergency essentials
Prepare an emergency kit
British Red Cross
Practical suggestions for supplies and personal needs. We prioritise what a household already has before buying more.
Security guidance · Recovery
Backing up your data
UK National Cyber Security Centre
Supports keeping recoverable copies of important files. Backups protect records; they cannot restore an external service during an outage.
Money guidance · Household buffer
Emergency savings: how much is enough?
MoneyHelper
Explains an affordable savings cushion alongside debt priorities. Savings buy time but cannot replace a salary indefinitely.
Education guidance · Published 2023
Guidance for generative AI in education and research
UNESCO
Puts learning, privacy and human agency at the centre. An AI answer is not proof that a child understands the subject.
Rights guidance · UK guidance under review
Decisions made without human involvement
Information Commissioner’s Office
Explains automated decisions and ways to challenge them. The ICO flags this guidance for review as UK law changes; check its current wording.
07 / 4 SOURCESMental wellbeing, medical progress and digital life
Guidance for coping with change, current medical research and the open questions around mind uploading. Future hopes are separate from treatments available today.
Wellbeing guidance · Coping with change
How to deal with change and uncertainty
NHS · Every Mind Matters
Supports manageable steps, routines and asking for help. Preparation should support wellbeing, not keep people in constant fear.
Medical research overview · Cancer research
Artificial Intelligence and Cancer
US National Cancer Institute
Describes AI in cancer biology, detection and treatment research. It does not establish that all cancers or all illnesses will be cured.
Clinical research report · October 2025
Eye prosthesis restores sight lost to macular degeneration
Stanford Medicine
Reports partial restoration of vision in a specific advanced eye condition. It is not evidence of a cure for every cause of blindness.
Researcher perspectives · September 2026
Can consciousness be uploaded to a machine?
MIT McGovern Institute
Discusses unresolved scientific and philosophical barriers. Brain mapping or a digital imitation does not demonstrate transfer of a human consciousness.
Country shading summarises disruption at monitored Cloudflare locations and active Google Cloud regions, not the availability of every service in a country. Missing or stale source coverage is shown as unknown. Red dots represent dated reviewed cases and automatically located news or official briefing reports. Automated reports stay at the smallest impact size and do not imply an ongoing attack; nationwide advisories have no invented location. Independent project. No affiliation with the organisations above. Assessments are kept separate from commercial product recommendations. Household guidance is currently UK-specific; international editions are planned. Contains NCSC information licensed under the Open Government Licence v3.0.