← all systems · see it in The System
Phishing
A form of social engineering and a scam in which attackers deceive people into revealing sensitive information or installing malware (src:wp:phishing); in practice an attack with a life cycle: infrastructure obtained, a phishing website configured (often from a kit), lures distributed, victims visiting, detection and mitigation, monetisation of what was stolen, and the site going offline (src:oest-2020-sunrise). Walked as security researchers describe it: its forms (bulk, spear, whaling, voice, SMS, QR, adversary-in-the-middle, lateral), its scale, its economics (kits, markets, the commons), its defences (filters, blocklists, takedown, training, phishing-resistant authentication) and its history since 1995.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | AOHell released 1994; phishing of AOL accounts reported 1995; term in print by March 1997 |
| born precision | year |
| boundary | Inside: one phishing attack's life cycle and the machinery it runs on: the domain and hosting, the kit or reverse proxy, the lure and its distribution, the counterfeit page, the victim's submission, the harvest and its route to the phisher, and the monetisation of what was taken. Outside: the people deceived (Human), the brands imitated and the organisations hit, the phisher as an actor (threat actor), the underground markets, and the defences that answer it (awareness, filters, blocklists, takedown, authentication, law). |
| nested level | Runs on people (the victim's senses and decisions, Human L5-L7) and on networks (email, SMS, telephone, the web, Internet services L6); a variant of social engineering and a tool of the threat actor; one of the commonest first steps into larger intrusions and frauds. |
| scale | technological |
Measures (39) published science
| Peak browser mitigation after detection | about 7 hours |
| Campaign duration, first to last victim (average) | 21 hours |
| Detection after first victim visit (average) | about 9 hours |
| Visitor to fraud yield (lower bound) | 7.42 % of visitors |
| Victims lured before detection | 62.73 % |
| Victim traffic after detection | 37.73 % |
| Share of victims taken by the top 10% of attacks | 89.13 % |
| Average date range of a campaign's compromised visitors | 13.55 days |
| Bulk lure click-through (as cited) | 5-8 % |
| Credential hand-over by targeted users (as cited) | 9 % |
| Lure to click to dangerous action (per hand-off yield) | sent to click 0.0567; click to dangerous action 0.7313 fraction |
| Time to employee reports of a campaign | about 10% within 5; 20% within 15; 30-40% within 30 minutes after delivery |
| Phishing site lifetime, free hosting, brand owner aware (mean / median) | 4.3 / 0 hours |
| Simulated phishing click rate, US health care | 14.2 (422,062 of 2,971,945); median 16.7 % |
| Odds of clicking after repeated campaigns (adjusted OR) | 0.511 (6-10 campaigns); 0.335 (more than 10) odds ratio |
| Phishing site lifetime, free hosting, brand owner missed (mean / median) | 114.7 / 29 hours |
| Rock-phish domain lifetime (mean / median) | 70.3 / 33 hours |
| Fast-flux domain lifetime (mean / median) | 96.1 / 25.5 hours |
| Fast-flux IP rotation | 5-10 addresses, changed as often as every 20 minutes IPs; minutes |
| Participants fooled by the best phishing site | more than 90 % |
| Participants who never looked at browser security cues | 23 (5 of 22) % |
| Overstatement of survey loss estimates (Herley and Florêncio) | up to 50 times |
| Potential phishing-kit victims, March 2016 to March 2017 | 12.4 million people |
| Hijacking odds ratio, phishing-kit victims against a random user | 463.4 (over 400x) odds ratio |
| Exposed phishing passwords matching the victim's current Google password | 24.8 % |
| Kits collecting geolocation / phone / user agent | 83 / 18 / 16 % |
| Gmail share of kit exfiltration addresses | 72.3 % |
| Distinct phishing kits behind a year's active attacks | 4,069 kits |
| MITM phishing domains / IPs on blocklists | 43.7 / 18.9 % |
| MITM phishing websites found in a year | 1,220 sites |
| MITM domains registered in the week of detection | 45.3 % |
| MITM site ready after its TLS certificate | all within one day; over 90% immediately time |
| Lateral phishers who compromise another employee | over 11 % |
| Lateral attacks in the hijacked account's working hours | over 80 % |
| Click-through: control / human experts / fully AI / AI with human in loop | 12 / 54 / 54 / 56 % |
| Profitability gain from AI for larger audiences | up to 50 times |
| Phishing attacks counted by APWG | 3.76 million (2024); 3.8 million (2025) unique reported phishing websites per year |
| FBI IC3 phishing and spoofing complaints, 2025 | 191,561; reported losses about US$215.8 million complaints; dollars |
| Phishing share of initial access (ENISA 2025, as cited) | about 60 % |
Open contentions (5) open contention
Where sources or readings disagree. Both stay on the map.
- C-663 Are shared resources always overused (the 'tragedy of the commons')?
booked: The map keeps both: the tragedy as a pattern, and users preventing it with their own rules, which is well supported across 91 studies.
the other reading: users close commons with rules (Ostrom) - C-1348 Does training people right after they click a fake phishing email help?
booked: The map records the large field study that found it made employees more likely to click again, and keeps the earlier, smaller studies that found it helped alongside.
the other reading: embedded training improves resilience - C-1752 Did human deception grow out of deceiving prey?
booked: The map books that suggestion as its authors', with the view that deception evolved to fool others of our kind.
the other reading: deception evolved for deceiving conspecifics (Machiavellian intelligence, Social Brain Hypothesis); the Foraging Brain Hypothesis without a deception link - C-1765 Do phishers keep evolving against defenders?
booked: The map books an arms race, and keeps the finding that phishing kits barely changed for years.
the other reading: a remarkable lack of external pressure: kit playbooks largely unchanged since the mid-2000s (Thomas et al. 2017) - C-1766 How much money does phishing really cost?
booked: The map keeps both reported losses and the estimate that losses are overstated and phishers earn little.
the other reading: phishing is lucrative (the 'easy money' view the paper argues against); phishing's losses are small in dollars and its cost lies in lost trust and email's usefulness
Partners and parts published science
Built from / builds (9)
- Phishing kits and phishing-as-a-service
- MITM (reverse-proxy) phishing toolkit
- Lure (email, text, call, QR code)
- Counterfeit login page
- Domains and hosting
- Exfiltration point (drop)
- Phishers against legitimate sites for a person's trust
- Phishers against claimants and verifiers for credentials
- Robocallers and call authentication competing for the called party's ear
Sits inside / holds (3)
Trades with (valence) (15)
- Authentication via sys:internet · Phishing mixed
- Human (Homo sapiens) via sys:electromagnetic-field · Phishing gain
- Internet via sys:internet · Phishing gain
- Large language model via sys:mental-representation · Phishing mixed
- Market and trade via sys:market · Phishing mixed
- Electronic mail via sys:internet · Phishing gain
- Computer security via sys:internet · Phishing loss
- Security operations center (SOC) via sys:internet · Phishing loss
- Threat actor via sys:market · Phishing gain
- Organization via sys:internet · Phishing gain
- Security awareness via sys:electromagnetic-field · Phishing gain
- Digital identity via sys:internet · Phishing gain
- Cybercrime and its underground economy via sys:internet · Phishing gain
- Computer law via sys:internet · Phishing loss
- Telephone via sys:electromagnetic-field · Phishing gain
expressed by (8)
- Templating / Standardization
- Amplification/Scale-Up (Manufacture & Distribution)
- Evade Detection
- Speed
- Precision Attack
- Subversion & Co-optation
- Leverage Host/Amplification
- Redundancy & Diversity
floor filled by (15)
- Threat actor
- Human (Homo sapiens)
- Internet
- Market and trade
- Electronic mail
- Organization
- Telephone network (circuit-switched carrier)
- Cybercrime and its underground economy
- Social engineering
- Phishing kits and phishing-as-a-service
- MITM (reverse-proxy) phishing toolkit
- Lure (email, text, call, QR code)
- Counterfeit login page
- Domains and hosting
- Exfiltration point (drop)
Kinds (12)
- Cybercrime and its underground economy
- Social engineering
- Typosquatting
- Whaling
- Spear phishing and whaling
- Adversary-in-the-middle phishing
- Business email compromise
- Lateral phishing
- Simulated phishing (phishing tests)
- Voice phishing (vishing)
- SMS phishing (smishing)
- QR code phishing (quishing)
Notes published science
- Protection is often worth less than it costs the people asked to do it: a minute a day spent checking web addresses would cost users about a hundred times more than all phishing losses, so ignoring the advice is rational (src:herley-2009-externalities); and a spam campaign burns more than ten dollars of other people's value for each dollar the spammer makes.
- In a 15-month study of 14,733 employees, a third clicked at least one fake phishing email, but the same people also worked as an alarm system: about 10% of their reports of a new phish came in within 5 minutes. The common industry fix, showing a training page to those who clicked, made them more likely to click again (src:lain-2022-phishing).
- In a 15-month study of 14,733 employees, the common practice of showing a training page to people who click a fake phish made them more likely to fall for phishing, not less; many said the page made them feel safe (src:lain-2022-phishing-org).
- Reed warblers do what security teams do: as cuckoos grew scarce at one English fen (from 24% of nests parasitised in 1985 to 1% in 2012), the birds dropped their guard, with egg rejection falling from 61% to 11% (src:thorogood-davies-2013-cuckoo). And a fully AI-written spear-phish now gets clicked as often as one written by human experts, 54% against 12% for generic phishing (src:heiding-2024-llm-phishing).
- In a hospital system with 19,500 staff, how recently someone had done the annual security course made no measurable difference to whether they clicked a fake phishing email; over half of the 'you clicked, here is your lesson' pages were closed within 10 seconds, and by month eight 56% of staff had clicked at least one test lure (src:ho-2025-phishing-training).
- Feeling protected lowers the guard: in a company of 14,733 people, those shown a training page after clicking went on to click more, and 43% of those who remembered it said it made them feel safe; in German hospitals, doctors with better IT tools showed less security awareness (src:lain-2022-phishing-organizations, src:kraushaar-2025-hospital-security).
- Staff make a fast alarm system: about 10% of reports of a phishing email arrived within 5 minutes of delivery, and people who got a thank-you style result back reported more (src:lain-2022-phishing-organizations). The WWII 'Loose lips sink ships' posters, by contrast, may have been more about stopping gloomy rumours than spies: the FBI had rounded up the key German agents in June 1941 (src:wp:loose-lips-sink-ships).
- Phishing may not pay. Herley and Florêncio argue it is a classic tragedy of the commons, like an open-access fishery: so many phishers graze the same pool that the average one earns about what he could legally, and the flood of phishing email is evidence of failure, not riches. The name fits better than its coiner knew (src:herley-florencio-2008-commons).
- A typical phishing campaign lives and dies in about 21 hours, and most of its victims arrive before anyone notices: detection comes about nine hours after the first victim, by which time 62.73% have already been lured. Yet a day after detection victims start rising again, because people open the old emails (src:oest-2020-sunrise).
- Phishing is an old trick in a new medium: an Australian katydid catches male cicadas by answering their songs with the female's wing-flick reply, and Photuris fireflies copy the female flashes of up to eleven prey species. Free phishing kits pull the same trick on their own users, quietly mailing every stolen password to the kit's author too (src:marshall-hill-2009-katydid, src:cova-2008-phishing-kits).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
phishing as security researchers describe it: the networks and brands it imitates, up to the phisher's return (reader's view)
- L1: The networks people trust and the brands they recognise: email, SMS, telephone, the web; banks, email and cloud providers, streaming services
- L2: Infrastructure and kit: a domain obtained (look-alike, compromised, free host, fast-flux) and a phishing site configured, often by installing a kit
- L3: Distribution: lures sent broadly or aimed (spear, whaling), by email, SMS, voice, QR code, or from a hijacked account
- L4: The counterfeit at the screen: impersonated sender, look-alike or homograph link, urgency, a copied or live-mirrored login page
- L5: The victim's judgement and act: click and submit credentials, or not (at least 7.42% of visitors go on to fraud)
- L6: The harvest: credentials, codes, session cookies and risk profiles collected and mailed to a drop, or relayed live
- L7: Monetisation: hijack, fraud, resale, further intrusion; at the scale of all phishers, a commons whose return falls to opportunity cost
the boardroom (shared floors)
- L4: The running campaign: kit, domain, lure and counterfeit page as one operation, from configuration to the last victim (on average 21 hours); a few campaigns take most victims
- L5: The recipient's judgement and act: perceive the lure, judge it, click and submit, or not. Runs on the person (Human's board: thoughts, feelings, emotions)
- L6: The harvest: credentials, codes, cookies and risk profiles collected and moved to the phisher's drop, or relayed live
- L7: The phisher's return: accounts hijacked, fraudulent transactions, resale, further intrusion or espionage; in open access, the average return driven down to the phisher's opportunity cost
way in: Current events
- L1: The news and calendar: a pandemic, a sales day, tax time
- L2: Read: campaign peaks checked against the timelines of pandemic measures and events
- L3: Themed: old schemes adapted to the event rather than reinvented
way in: Detection met on the way: filters, blocklists, crawlers, takedown notices
- L1: The anti-phishing ecosystem: clearinghouses, crawlers, browser blocklists, brand owners' takedown requests
- L2: Felt as a gap closing: blocklists are reactive; one changed character makes a URL unknown; cloaking tells crawlers from victims
- L3: Mitigation lands: warnings halve effectiveness within two hours of detection; MITM sites mostly escape (43.7% of domains listed)
way in: Domains and hosting
- L1: The domain-name and hosting world: free web-hosting, compromised machines, newly registered and look-alike domains, TLS certificates
- L2: Obtained: a look-alike or meaningless domain registered (45.3% of MITM domains in the week they were detected), a legitimate site compromised, or a fast-flux name resolving to 5-10 changing IPs
- L3: Configured: the kit installed on the infrastructure; a MITM site answers within a day of its certificate
way in: Emotion: urgency and threat
- L1: Urgency and threats written into the message (closing or seizing an account)
- L2: Threat and urgency cues are looked at first
- L3: Threatening language lowers rated trust
way in: Experience: simulated campaigns and training
- L1: Simulated phishing sent by the organisation; training material shown after a click
- L2: Repeated exposure: more campaigns, lower odds of clicking (aOR 0.335 after more than ten)
- L3: Embedded training can make people more susceptible
way in: Hearing: a voice on the phone
- L1: A telephone or VoIP call, often automated text-to-speech
- L2: A spoofed caller number and interactive voice response
- L3: Trust: voice telephony trusted more than email
way in: Relay: the real service's live session
- L1: The legitimate service answering a login that the victim thinks is direct
- L2: The reverse proxy forwards requests and responses, valid TLS certificate in place
- L3: Credentials, one-time codes and the session cookie (a bearer secret) extracted in transit
way in: Sight: the lure and the page on a screen
- L1: The lure in the inbox or on the screen: the From: header the recipient sees, not the unchecked envelope sender
- L2: The eyes fixate on threat and urgency before misspellings; indicators looked at often but briefly; a QR code's destination unreadable
- L3: Cues read or missed: 23% never looked at the address bar or security indicators; homograph domains are visually identical
- L4: Judged: emails with misspellings or threats rated least trustworthy, yet good phishing sites fooled 90%
way in: Submissions: what the victim types
- L1: The victim's username, password and codes, posted from the counterfeit page
- L2: Received by the kit's script on the phishing host
- L3: Enriched with the victim's risk profile: geolocation (83% of kits), phone (18%), user agent (16%)
- L4: Packed into a message template and mailed with PHP's mail()
- L5: Sent to the drop: Gmail 72.3% of kit exfiltration addresses; free kits copy it covertly to their authors
way in: Targets: address lists, profiles, hijacked contacts
- L1: People reachable by email, text and phone, and their public traces
- L2: Collected: bulk lists for general phishing; personal or organisational information for spear phishing; a hijacked account's contacts for lateral phishing
- L3: Chosen: sent broadly, or aimed at a person, a group, or senior leaders (whaling)
way in: The imitated brand: its look and its live pages
- L1: The impersonated brand's real website and resources (images, scripts); banks, email and cloud providers, streaming services are common targets
- L2: Copied: static copies fetched from the target website, or mirrored live by a reverse proxy; many phishing pages still request resources from the site they impersonate
- L3: A counterfeit page: a 'perfect' copy in MITM kits, branded templates in phishing-as-a-service
way in: The market: what stolen access is worth
- L1: A black market with specialisations: phishing software for payment, credentials and accounts for sale
- L2: Offers and prices read; services sold by those who found phishing unprofitable
- L3: Perceived return: survey estimates overstated up to fifty-fold keep new entrants coming
way in: Thought: knowing what the indicators mean
- L1: The browser's chrome and security indicators, and page content designers (or phishers) can fill freely
- L2: Knowledge: many users do not understand security indicators
- L3: Content-only judgement: 5 of 22 judged by the page content alone
way in: Tools: kits, platforms, language models
- L1: Kits and phishing-as-a-service advertised, sold or given away; language models
- L2: Taken up: complete phishing sites in an easy-to-deploy format, configured by changing one line
- L3: Playbooks largely unchanged since the mid-2000s; AI-written spear phishing on par with experts
way out: Distribution: lures sent
- L1: Lures in inboxes and on phones
- L3: Distributed by email (dominant, helped by bulk-messaging services), SMS, calls, QR stickers
- L4: The campaign launched
way out: Further intrusion or espionage
- L1: Further systems, records and people
- L4: Phished subcontractor account used against Target (110 million records, 2013); SecurID data taken from RSA (2011); political campaigns spear phished (2016)
- L7: Access used for a larger objective
way out: Lateral: the hijacked account sends the next lures
- L1: Colleagues' inboxes
- L3: Shared-document and account-warning lures sent to colleagues, over 80% in the account's working hours
- L6: A harvested enterprise account
way out: Monetisation: accounts hijacked, fraud committed
- L1: Fraudulent transactions; accounts taken over (phishing victims 400x likelier to be hijacked)
- L6: Stolen credentials tested against the platforms; risk profiles used to impersonate the victim
- L7: Use the take
way out: Resale and the kit author's cut
- L1: Accounts and credentials sold on darknet markets
- L6: Backdoored free kits send the take to their authors
- L7: Sell rather than use
way out: The counterfeit page served
- L1: The page on the victim's screen
- L2: Served to victims, cloaked from crawlers; MITM kits mirror the live site with a valid certificate
- L4: Site live