About CyberView
A live, verified picture of cyber threats worldwide for defenders: what happened, what is being exploited, and the outages around it, updated around the clock.
Free, non-commercial and open source.
What CyberView is
CyberView watches the world's public reporting on cyber security and turns it into one place a defender can use. Scripts collect from 280 government, vendor and news feeds in more than 30 languages, with regional sources for Africa, the Middle East, Turkey and Latin America, standing news searches in local languages and several open databases every few minutes; CyberView AI (Claude models by Anthropic) researches what they find, writes it up with its sources, and a second AI agent checks it before it counts as reviewed. Nothing is drawn as a fact until it has been read from a real source.
It keeps growing. New sources, views and analyses are added continuously, and this page is generated from the same list the site is built from, so it always describes what CyberView does today: 559 events on file, 430,470 CVEs indexed, 41 power grid sources connected.
- Seven-day globe of cyber events: Every checked cyber event of the last seven days on a 3D globe, by type (shape and colour-blind-safe colour) and severity, with what happened, why it matters, what to do and the sources. Arcs connect events only where a source states the link, or where a government or the investigating firm made an attribution, and say who said so.
- Timeline and Correlation: The same events by day and attack surface, and as a correlation graph of shared actors, products and techniques, each with the question it answers and its caveats. Sector and attack surface come from fixed lists (the US CISA critical infrastructure sectors and a few more), each stated by the source or inferred by CyberView AI (a Claude model); events not yet classified are placed by keyword, and each view counts and labels them as keyword-assigned.
- Continuous research loop: A loop runs every few minutes: it collects leads, groups them into stories ranked by how many independent publishers report them, sends researchers to new stories and reviewers to new events in parallel, and publishes. The target is under 15 minutes from a lead to a published, reviewed event. Routine bulletins and single unconfirmed headlines are closed by written rules, and unconfirmed reports appear only in a labelled 'being checked' list.
- Official breach notices: Breach notices that organisations file with regulators are read straight from the registers that publish them (the California, Washington and Iowa attorneys general, the UK Information Commissioner's enforcement actions, and US SEC filings for material cybersecurity incidents once a contact address is set). Each new notice is a lead from a government source: a researcher reads it, cites the regulator's own page, and adds it to the map or to the event it confirms. Notices of the same breach on several registers are counted as one story.
- Events that keep up with the story: Published events are updated when new reporting adds a confirmation, a figure, an attribution or a patch, and go back to the second review after each change; developing stories and attackers' claims are followed up on a schedule. Corrections are logged publicly.
- Vulnerabilities: Known-exploited flaws ranked for patching with US federal deadlines and fix steps from the vendor's own advisory; 430,470 CVEs since 1999 to look up and browse, each with the CyberView rating (the CVSS base score adjusted for exploitation evidence with CVSS's own Exploit Code Maturity metric, on the CVSS scale and bands) and an urgency priority that follows CISA's SSVC and BOD 26-04; the vendor's own advisory joined to each CVE (twelve vendors: severity and fixed versions as the vendor states them, and its own word when a flaw is exploited in the wild, which raises the rating); your own software list matched by version, in your browser, against NVD's affected-version ranges for every CVE that has them (not just the known-exploited ones) and never uploaded; and hunt guidance. Exploitation analyses by CyberView AI explain how a flaw is used and how to fix it, without exploit code.
- Industrial (ICS/OT) advisories: An Industrial view on the Vulnerabilities page with every industrial control system and operational technology advisory of the last year from CISA (ICS and medical devices), Siemens, Schneider Electric, ABB, Phoenix Contact, CERT@VDE, Moxa and Mitsubishi Electric, read by a script from their own machine-readable feeds: affected products, CVEs, CVSS, what the advisory and CISA's known-exploited catalog say about exploitation, the critical infrastructure sectors, and a link to the advisory with its mitigations. Each CVE elsewhere on the page shows the industrial advisories that name it. Exploited and critical advisories, and high-scoring ones in energy, water, health and other lifeline sectors, are handed to CyberView AI to research.
- Vulnerability scanner for your own network: A command-line tool to download and run against systems you own or are authorized to test. It reads the versions your services state about themselves (SSH banners, web server headers, mail greetings, TLS certificates, and optional read-only DNS, NTP and SNMP service checks), matches them against CyberView's known-exploited flaws with the same code as the Vulnerabilities page, and ranks what to patch first by the CyberView rating, with every point listed, the fix and a confidence for each finding, exported as JSON, CSV, SARIF and a software list. It can find the hosts for you within the scope you attest to: sweep your ranges, resolve and expand your own domains from Certificate Transparency, import an asset list, or read your own read-only cloud inventory; every host found is re-checked against your scope before it is scanned. It can scan every TCP port on your own private ranges. It detects and never attacks: it will not start without a scope file and the operator's recorded attestation; it needs an extra confirmation for public addresses and keeps a hard cap on how much public space it will ever touch; its UDP checks send a single benign query per service with only the standard read community; it never logs in, guesses passwords or community strings, runs your cloud CLI or reads your keys; it keeps polite limits that can only be lowered; it names itself in its requests, works offline from a downloaded copy of CyberView's data, and never uploads results. No AI model runs in it.
- Power grid outages: A Grid view that sets everything else aside and shows only electrical grid outages from 41 grid operators and data sources (utilities, system operators, and the statewide or national counts that government emergency agencies publish, such as Pennsylvania's, Maryland's and Chile's, plus Japan's supply reserve per area from OCCTO): every outage of 100 or more customers, larger incidents tracked from start to end, load shedding and grid emergencies, which outages are unusual for the hour, which are confirmed by internet measurements, disaster alerts, demand data or news, and 30-day trends. A third tier lists power outages reported in the news in many languages, grouped by place and time with each publisher named and linked, found by a script and labelled as not checked by CyberView until CyberView AI's outage desk has checked them; they are never counted as incidents. CyberView AI checks the cause of big or corroborated incidents; an outage is never called an attack unless an official source says so. A rule-based audit checks every grid source each cycle (failing or frozen feeds, sudden falls in customers, misplaced points), and the status page lists its flags.
- Context layers: Internet disruptions (IODA) and disaster alerts (GDACS) on the globe, drawn in neutral ink and always labelled as context, not cyber events; a strong internet disruption shows the outage desk's check when there is one.
- Scam warnings by country: Official warnings about scams and phishing from government bodies, police forces, regulators and national cyber security centres in several countries and languages, read every two hours from the issuers' own feeds by a script with no AI model, and listed by country in the map page's Insights tab with the issuer named and a link only to the issuer's own page. The scam type (impersonation, investment, delivery, bank or SMS, tech support, romance, job, crypto) is assigned by a fixed keyword list and labelled so. They are warnings, not cyber events: never drawn on the globe or counted as events; a warning that describes a cyber incident is passed to the research loop as a lead.
- Outage checks: An outage desk inside the research loop, where CyberView AI (a Claude model) checks outage reports that no one had checked: power outages reported in the news that no operator feed covers, country-level internet disruptions measured by IODA, and major outages on providers' own status pages. Each is read against the operator's, provider's or government's own page and credible reporting and marked confirmed (naming who confirmed it, with figures, times and cause only as a source states them), not confirmed (with the reason), or, when a source links it to a cyberattack, handed to the research loop as a lead, never published as an attack by itself. The Grid view and the internet disruption layer show the result next to the report, and reports not checked yet stay labelled as unchecked. A confirmed internet or service outage can become an outage event, with a second review like every other. The service outages come from the providers' own status pages (cloud, CDN, DNS, SaaS, payment and telecom), read by a script with no AI model; major or long incidents are listed under Service status in the map's layers panel, each labelled as the provider's own status page and not a cyber event, and the outage desk checks the biggest providers first.
- Leak-site claims checked: A claims desk inside the research loop, where CyberView AI (a Claude model) checks the claims ransomware and extortion groups post on their leak sites, most notable victim first: hospitals and health, government, energy and other critical sectors, large companies, victims other reports already name, and groups whose claims have been confirmed before. For each claim it looks for the organisation's own statement, a regulator filing or a government notice, then credible reporting, and records one result: confirmed by the victim (an event with a second review), reported by two publishers but not confirmed (kept as an unverified claim, severity low and hidden by default), denied, nothing beyond the leak site (checked again later) or victim not identifiable. Events show the check as "Checked by CyberView AI (a Claude model)". Leak sites are never cited or linked, only the group's name and the claim's date are recorded, and stolen data is never opened or described.
- Brief, patterns and trends: A daily brief and a pattern analysis written by CyberView AI from the week's events, and week-over-week trends by category, sector, country, actor and product.
- Threat actor profiles: One profile per group named in CyberView's checked events, put together by a script, not written: the events a government or the investigating firm attributed to it (each saying who said so), the events whose reporting names it, and, kept apart and labelled as unconfirmed, what it only claimed on its leak site. Other names are joined only where MITRE ATT&CK lists them as aliases or an event's sources name them as one group, and each says which. Each profile lists sectors and countries targeted, CVEs and government advisories in its events, and links to them in the archive.
- Archive: Every event CyberView has ever recorded (559 so far), searchable, with RSS and JSON feeds. Events that research finds after the fact (they happened more than a day before CyberView filed them) are routed here under Recently filed, with the day they happened and when they were filed, so the map's Just added list only ever shows new events.
- Weekly tabletop exercise: Each week one verified event becomes a ready-to-run exercise with timed updates, discussion questions and what good looks like; the scenario is invented for practice and says so.
- Community: Defenders' tips and questions, reviewed by fixed rules and then the CyberView AI moderator before they appear; never used as a source for an event.
- Data CyberView would like to use: Every power grid and cyber security source CyberView would like to show but may not yet, because its terms restrict reuse, its robots.txt disallows the data, its site blocks automated readers, it needs a key or it could not be reached, each with the reason, the contact its operator publishes, a short request written for that operator and the order to ask in. CyberView never works around a block; a source is connected only once it is allowed.
- Open about how it runs: A status page that answers first whether the site is healthy right now, followed by how the data flows (sources, leads, stories, closed by rule, read by AI agents, published, second-reviewed and on the map, each with its count for the last 24 hours and 7 days and the share it passes on) and where the data comes from (events of the week by source tier, collector, region and language, and the publishers they cite most) and what needs attention, with every detail one click away: every routine, when it last ran and what it found; an hourly pipeline check, a script with no AI model, that measures every stage from the collectors to the deploy against its limit, restarts a stage that has stopped, and checks the data itself (records vanishing from the status file, events recorded twice, events waiting over two hours for their second review, lead files or the data growing abnormally, feeds and collectors that keep failing, which are asked less often and proposed for the watch list instead of being dropped silently), listing every problem under what needs attention and opening one public GitHub issue per problem that lasts, closed again when it is healthy; an outside guardian that checks the live site and every stage every two hours and restarts what has stopped; link and security checks; an improver that tunes cost and quality within written limits; and an efficiency report, worked out by a script with no AI model at every build: usage and idle runs per routine, leads and how many became events per feed, coverage by region, language and sector, and corrections by week, each saying how many days of data it rests on. The code and the job instructions are public.
How it is made
- Collection: scripts read more than 220 government, vendor and news feeds in 28 languages, ask two independent news indexes (Bing News and GDELT) standing questions in 13 languages, keeping only links to the publishers' own pages, and read primary databases: CISA's known-exploited list, ENISA's EU Vulnerability Database, NIST NVD, FIRST EPSS, Have I Been Pwned, cloud providers' status pages and leak-site tracking. No AI is involved in this step, and no single search engine decides what CyberView sees.
- Corroboration: leads about the same thing are grouped into one story, and stories reported by more independent publishers are researched first. Leak-site listings are attackers' claims: they are never published as facts.
- Research: CyberView AI (Claude models by Anthropic, working as scheduled jobs) turns leads into events and writes every summary, brief, analysis and exercise on the site. They work under written rules that are published in full: facts only from pages actually read, original government and vendor documents preferred, specific source links required, outages never called attacks without an official source, attacker claims marked unverified.
- Second review: every new or changed event is re-read against its sources by a different AI agent from the one that wrote it, which passes it, corrects it or flags it. Each event shows whether that review has happened.
- Checking: a verification job re-checks recent events against their sources, the verify desk answers reader reports, an outside guardian checks the live site every two hours, and every build validates every record. Corrections are logged publicly.
- Severity follows exploitation: known-exploited flaws, public exploit code or an official "patch immediately" in widely used or internet-facing software are rated high.
- Links and attribution on the globe: arcs connect events only where a source states the link (for example a breached supplier and a customer hit through it). Attribution arcs are drawn only when a government body or the investigating firm made the attribution, and name who made it; CyberView does not attribute attacks itself.
- Vulnerabilities: the Vulnerabilities page ranks exploited flaws by an open score, compares the software versions you list against NIST NVD's affected ranges inside your browser, and shows check and fix steps that an AI job takes from the vendor's own advisory. Version data can lag or be incomplete: confirm against the vendor advisory before you rely on a "not affected".
- Community feed: visitors' posts are reviewed before they appear, first by fixed rules and then by the CyberView AI moderator (a Claude model). They are the authors' own words, not CyberView's reporting, and are never used as a source for an event; a credible tip is passed to the research jobs to verify from real sources.
- Tabletop exercises: each week an AI job turns one verified event into a practice exercise. The real facts carry their sources; the scenario for your organisation is invented for practice and says so.
- Every routine and its last run are shown on the status page; the full history is in the archive and the public repository.
AI-assisted research can make mistakes. Read information, not advice before relying on an entry.
Sources and licences
- Event summaries and structured data written for CyberView: you may reuse them with attribution ("CyberView") and a link to the event. Each event links to the articles and advisories it is based on; those remain the property of their publishers.
- CISA Known Exploited Vulnerabilities catalog: US government work, public domain.
- EPSS scores: FIRST.org; see EPSS usage guidance.
- CVSS data: NIST National Vulnerability Database. This product uses data from the NVD API but is not endorsed or certified by the NVD.
- Internet disruption signals: IODA, Internet Intelligence Lab, Georgia Institute of Technology.
- Power grid data: published by each grid operator under its own terms; CyberView uses only sources whose terms and robots.txt allow it (see data CyberView would like to use). Credits the licences require:
- Grid Alert text from the AESO (www.aeso.ca), unmodified, non-commercial use. Copyright © 2016 – 2026 Alberta Electric System Operator. All rights reserved.
- Copyright © 2004-2026 Independent Electricity System Operator, all rights reserved. This information is subject to the Terms of Use set out in the IESO’s website (www.ieso.ca)
- SSEN Distribution, Real Time Outage Dataset (realtime_outage_dataset); 2026-11-22. CC BY 4.0
- Supported by NGED Open Data
- Contains BMRS data © Elexon Limited copyright and database right 2026
- Liander, Storingsdata (actuele storingen). CC BY 4.0
- Source: OCCTO (電力広域的運営推進機関), 広域予備率Web公表システム. Changes: CyberView reads the current half hour's wide-area reserve rate per area and turns it into a level (5%, 3%, 1%).
- Source: Fingrid
- Source: RTE, Ecowatt
- Source: Korea Power Exchange (KPX)
- Source: California ISO (CAISO), www.caiso.com
- Taiwan Power Company 2026, 台灣電力公司今日系統供需狀況 (Taipower today's system supply and demand). The Open Data is made available to the public under the Open Government Data License, User can make use of it when complying to the condition and obligation of its terms. Open Government Data License: https://data.gov.tw/license
- Source: ONS - Operador Nacional do Sistema Elétrico, dataset Interrupção de Carga (CC-BY). Changes: CyberView keeps only the last 7 days and restates each row as MW of load interrupted per state.
- Disaster alerts: GDACS, a cooperation framework of the United Nations and the European Commission.
- Leak-site claims: listings tracked by ransomware.live. Organisation names only; never stolen data.
- Map: Natural Earth (public domain) via world-atlas; d3 (ISC licence) and topojson-client (ISC licence). Accounts use supabase-js (MIT licence).
- Sources CyberView may not use yet: CyberView uses a source only when its terms and robots.txt allow it, and never works around a block. Power companies and grid operators whose data it would like to show, why it cannot yet, and the request it sends them are listed on Data CyberView would like to use.
- Code: CyberView's own code is released under the MIT licence.
Corrections and removal requests
If an entry is wrong, use Report an error on the event, or open an issue in the repository. The verification job checks every report against the sources; when an original contradicts an entry, the entry is fixed and the correction is logged publicly with what changed.
If you are an organisation or person named in an entry and believe it is inaccurate, unfair, infringes copyright or breaches your privacy, contact us privately (below) with the event link and what is wrong. We review requests promptly, correct or remove content that is inaccurate or unlawful, and tell you the outcome. Attacker claims the named organisation has not confirmed are labelled as such and hidden by default; an organisation's denial is recorded when it is published.
Private contact: cyberviewthreats@gmail.com
Security disclosure
If you find a security problem in CyberView, please report it privately through GitHub's private vulnerability reporting, not in a public issue. Good-faith research that respects these rules is welcome: do not access other people's accounts or data, degrade the service, or use social engineering; give us reasonable time to fix the issue before disclosure. We do not pay bounties. Machine-readable details: security.txt.
Accessibility
We aim to meet WCAG 2.2 level AA. Event types are shown by both colour and shape using a colour-blind-safe palette, the map and lists work with a keyboard (press ? on the map for shortcuts), motion respects your reduced-motion setting, and every map event is also listed as text in the feed. Known limitation: the map and charts are visual summaries; the feed, archive and exports carry the same information as text. If something is hard to use, please tell us.