← all systems · see it in The System
Social engineering
In information security, the use of psychological pressure to influence people to perform actions or divulge confidential information (src:wp:social-engineering-security); a kind of psychological attack that persuades a victim to act as intended by an attacker, exploiting weaknesses in human cognition (src:montanez-2020-cognition). Walked as the security and psychology literature describes it: the approach (reconnaissance, pretext, lure on a channel: email, text, phone, QR code, a dropped drive, a person at the door), the persuasion principles it leans on, the target's cognition it works through, what the target does (discloses, runs, pays, admits) or does instead (suspects, verifies, reports), and how people and organisations resist it (awareness, warnings, crowd reporting, phishing-resistant authentication, cooling-off periods, law).
Properties published scienceour proposal
Sourced fields, and our own readings.
| aliases | sys:social-engineering-security, Social engineering security |
| born at | the shell game, at least ancient Greece; the 'confidence man' named in 1849; phishing of AOL accounts reported in 1995 |
| born precision | millennia (kind); year (security form) |
| boundary | Inside: the approach (reconnaissance, pretext, the lure on a channel), the persuasion it carries, the target's cognition it works through, and the act it asks for (disclose, run, pay, admit) or the resistance it meets in the target (suspicion, verification, report). Outside: the attacker as an actor (threat actor), the organisation and its assets, the awareness programmes, filters, authentication and laws that answer it, and the larger intrusion or fraud it opens. |
| nested level | Runs on people: on the target's cognition (Human L5-L6) and on the attacker's craft (threat actor's lure outlet); one of the threat actor's ways in and the inside face of many intrusions; a modern branch of the confidence trick. |
| scale | social |
Measures (25) published science
| Short con duration | minutes, possibly seconds time |
| Long con duration | several days or weeks time |
| Time to employee reports of a phishing campaign | about 10% within 5; 20% within 15; 30-40% within 30 minutes after delivery |
| Knowledge retention after phishing training (as cited) | a few days to a few months at most time |
| Study period over which young users' susceptibility declined | 21 days |
| Simulated phishes clicked (one company, 15 months) | 6,680 of 117,864 (5.67%) emails |
| Employees clicking at least one simulated phish in 15 months | 4,729 of 14,733 (32.10%) employees |
| Dangerous actions (credentials, macros) as share of clicks | 73.13 % |
| Reports received in 15 months (participants and others) | 18,476 emails |
| Employees doing at least one dangerous action in 15 months | 3,747 of 14,733 (25.43%) employees |
| Users clicking at least one of 21 daily spear-phishing emails | 43.3 % |
| Spear-phishing click-through by author (control, human expert, AI, AI with human) | 12 / 54 / 54 / 56 % |
| AI-gathered target profiles accurate and useful | 88 % |
| Office keys surrendered to a stranger (no intervention vs intervention) | 62.5 vs 37.0 % |
| Lie-truth judgment accuracy without aids | 54 (47 of lies, 61 of truths) % correct |
| USB drives picked up / called home (campus drop, 297 drives) | 98 / 45 % |
| Lateral phishers compromising at least one more account | at least 11 % of attackers |
| Organisations with a lateral phishing incident in seven months | 14 % of sampled organisations |
| Lateral attacks in the hijacked account's working hours | over 80 % |
| Phishing emails clicked in health care organisations (as cited) | 14.2 % on average |
| Phishing attacks counted by APWG | 3.8 million (2025); 3.76 million (2024) unique reported sites |
| Business email compromise reported losses (FBI IC3, 2025) | 24,768 complaints; $3,046,598,558 complaints; US$ |
| Text-message scam losses reported to the FTC (2024) | 470 million US$ |
| Malicious domains registered the day after the Equifax help site | 194 domains |
| Conversion funnel, lure to dangerous act (per hand-off yield) | sent to click 0.0567; click to dangerous action 0.7313 fraction |
Open contentions (3) open contention
Where sources or readings disagree. Both stay on the map.
- 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: training reduces susceptibility (Bullee et al. 2015: key hand-overs 62.5% to 37.0%; young users' decline in Lin et al. 2019); the measured benefit comes mainly from the periodic reminder the exercise creates, not the material after a click (a 2024 field study, as summarised in src:wp:phishing); training helps only when it reaches habit (Montañez et al. 2020) - C-1694 Does pretending to be an authority make scams work better?
booked: The map lists authority as a persuasion principle and keeps the field experiment where the cue had no significant effect.
the other reading: in a field experiment the authority cue had no significant effect on handing over office keys (Bullee et al. 2015); authority and scarcity messages are judged more suspicious than liking ones (Lawson et al. 2018, as cited in Montañez et al.) - C-1733 Is social engineering as old as cons, or a computer-era practice?
booked: The map dates it with the confidence trick, as old as the shell game, and keeps the information-security start alongside.
the other reading: rung 12: the term in its information-security sense, phishing (from 1995) and the research field belong to computing
Partners and parts published science
Built from / builds (5)
Sits inside / holds (4)
Trades with (valence) (11)
- Human (Homo sapiens) via sys:sensory-environment · Social engineering gain
- Law (rules and law) via sys:human-society · Social engineering loss
- Large language model via sys:written-record · Social engineering gain
- Computer security via sys:electromagnetic-field · Social engineering loss
- Security operations center (SOC) via sys:human-society · Social engineering loss
- Threat actor via sys:internet · Social engineering gain
- Insider threat via sys:organization · Social engineering gain
- Organization via sys:human-society · Social engineering gain
- Penetration testing via sys:organization · Social engineering even
- Security awareness via sys:human-society · Social engineering loss
- Digital identity via sys:internet · Social engineering gain
expressed by (12)
- Templating / Standardization
- Amplification/Scale-Up (Manufacture & Distribution)
- Sensing/Scanning
- Mapping/Modeling
- Evade Detection
- Overcome Barrier
- Speed
- Precision Attack
- Ranged Attack
- Subversion & Co-optation
- Prepositioning & Latency
- Leverage Host/Amplification
floor filled by (14)
- Phishing
- Human (Homo sapiens)
- Large language model
- Human brain
- Emotion (feelings)
- Electronic mail
- Threat actor
- Persuasion
- Telephone network (circuit-switched carrier)
- Cybercrime and its underground economy
- Pretext
- Phishing kits and phishing-as-a-service
- Security awareness
- Email spoofing
Kinds (18)
- Phishing
- Confidence trick
- Penetration testing
- Pretext
- Spear phishing and whaling
- QR code phishing (quishing)
- Adversary-in-the-middle phishing
- Baiting
- Tailgating (piggybacking)
- Watering hole attack
- Quid pro quo
- Scareware
- Ad phishing
- Business email compromise
- Lateral phishing
- AI-automated spear phishing
- Smishing (SMS lures)
- Vishing (voice lures)
Notes published science
- The better the spam filter, the more dangerous the email that slips past it: people spot malicious emails worst when they are rare (1% of the stream in one experiment), so the few that get through are more likely to work (src:montanez-2020-cognition).
- 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).
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.
social engineering as security and psychology researchers describe it: people and the channels that reach them, up to the attacker's objective (reader's view)
- L1: People and the channels that reach them: inboxes, phones, text messages, websites they visit, premises, found objects; and their public traces
- L2: The lure delivered: an email, a text, a call, a QR code, a dropped drive, a person at the door (general phishing sent broadly; spear phishing aimed)
- L3: Pretext and impersonation: an invented scenario, a spoofed sender or caller ID, a cloned page, a hijacked colleague's account
- L4: Persuasion: influence principles and urgency carried by the message
- L5: The target's cognition: perception, working memory, decision making and action, with short- and long-term factors
- L6: Compliance or resistance: the target acts as intended (disclose, run, pay, admit) or suspects, verifies, reports
- L7: The attacker's objective: information, money, access; often one step in a larger scheme
the boardroom (shared floors)
- L4: The crafted approach: a pretext and an impersonated sender carrying influence principles (liking, reciprocity, social proof, consistency, authority, scarcity, unity; distraction, herd, need and greed, time) and timed for a manufactured crisis (the 'hurrah')
- L5: The target's cognition: perception, working memory, decision making and action on a foundation of long-term memory, moved by short-term factors (workload, stress, vigilance) and long-term ones (expertise, age, culture)
- L6: Persuaded or not: the target acts as the attacker intended (clicks, discloses, runs, pays, opens the door), or suspects, verifies and reports
- L7: The attacker's objective: information, money or system access, usually one step in a larger fraud or intrusion
way in: Defences met on the way
- L1: Content and reputation filters, sender authentication (SPF, DKIM, DMARC), browser and DNS blocklists
- L2: Lures blocked, or let through in the gap before a site is listed
- L3: Evasion: 31% of lateral phishers invest effort in evading detection; spoofed caller ID impedes detection
way in: Domain binding: the authenticator's own check
- L1: The relying party's actual domain, which a look-alike page cannot share
- L2: The authenticator answers a challenge bound to that domain (WebAuthn, passkeys)
- L3: A look-alike domain gets no usable response; one-time codes, by contrast, can be relayed
way in: Emotion: the lure's charge
- L1: The emotional charge the lure carries: urgency, fear of loss, need and greed, kindness
- L3: Background anxiety (the pandemic) raises the chance attacks succeed
- L4: Acute stress narrows attention (attentional tunneling): the message text, not the address or the warning
way in: Experience: what memory holds of lures and training
- L1: Past encounters with lures and training, held in long-term memory
- L3: Narrow security expertise helps; general computer expertise may not; overconfidence
- L4: Age: young users' susceptibility fell over 21 days of lures, older users' did not; older age with lower working memory raises it
way in: Hearing: a voice on the phone
- L1: A voice on the line, live or synthetic; spoofed caller ID; cloned voices of a trusted person
- L2: Heard: automated text-to-speech calls direct victims to a number; live callers pose as bank, police or provider
- L3: Trust in the telephone, greater than in email
- L4: Judged in real time: audible lies are judged more accurately than visible ones, yet partners are regarded as honest
way in: Hijacked accounts and their trust
- L1: A hijacked enterprise account: its mailbox, contacts and the implicit trust colleagues give it
- L2: Account taken over with phished credentials; the access may be used to spear phish others inside
- L3: Internal lures: a shared-document pretext or a fake account warning, sent during the account's working hours
way in: Prevalence: how often lures arrive
- L1: The stream of messages, malicious ones at 20, 5 or 1% (one experiment's groups)
- L3: Expectation: where attacks are expected the attacker is at a disadvantage
- L4: Detection accuracy lowest when malicious emails are rarest
way in: Reconnaissance: the target's digital footprint
- L1: The target's digital footprint: public profiles, posts, organisational pages
- L2: Reconnaissance by scraping and searching, by hand or by an AI tool (88% of AI-gathered profiles accurate and useful)
- L3: A profile for the pretext: personal facts (date of birth, account numbers, last bill) and a vulnerability profile
way in: Salient events and announcements
- L1: Public events: a pandemic, a data breach, government announcements, tax season
- L2: Watched by attackers and acted on fast: 194 look-alike domains registered the day after the Equifax help site went up
- L3: A lure theme fitted to the event and timed to announcements
way in: Sight: the lure on a screen or page
- L1: The lure as seen: text, logos, a display name, links, QR codes; email's protocols carry no sender authentication
- L2: Read by eye: perception converts what the senses sample into neural codes (Human sight channel)
- L3: Superficial cues read as legitimacy: professional appearance, logos, name spoofing (visual deception)
- L4: Attention: urgency cues crowd out deception cues such as the sender's address or spelling
way in: The physical world: drives, doors, visitors
- L1: Things and people met in person: drives left in bathrooms, lifts and car parks with curious labels; a courier at the door
- L2: Picked up, seen, met face to face: 98% of 297 dropped drives were picked up
- L3: Curiosity or greed (baiting), helpfulness (holding the door)
way in: The target's replies
- L1: Replies: a recipient questions the email's validity; a voice on the phone hesitates
- L2: Read by the attacker: 107 hijacked-account attackers received at least one reply
- L3: A reassuring follow-up ('It's safe to open'): 25% of reply-receiving attackers sent one
way in: Thought: the target's own reasoning about the request
- L1: Where the target's thoughts come from is contested (C-413); here, reasoning about whether the request is real
- L4: Decision making driven two ways: fast heuristics and biases, or conscious, controlled reasoning
way in: Tools and services: kits, platforms, voices, language models
- L1: Phishing kits, phishing-as-a-service platforms (Darcula impersonated more than 200 organisations), text-to-speech, language models
- L2: Kits clone login pages automatically and distribute links through templated infrastructure
- L3: Content made at scale: personalised emails on par with human experts; audio deepfakes of a trusted voice
way in: Training and simulated phishing
- L1: Awareness interventions; simulated phishing with training shown after a click
- L2: The periodic reminder the exercise creates (more than the material after a click)
- L3: Learning that reaches habit guides safer behaviour largely unconsciously
- L4: Felt safety: 43% of trained respondents said the training page made them feel safe
way in: Warnings on the lure
- L1: Warnings on top of suspicious emails; browser warnings for blocklisted sites
- L2: Seen on top of the email, short or detailed
- L3: Effective; detailed warnings not significantly better than short ones
way in: Workload: the working day
- L1: The working day: large volumes of email, tasks, deadlines
- L3: Workload: the only factor that predicted clicking among hospital employees
- L4: Vigilance declines the longer a task runs; a high email load triggers automatic responses
way out: Admission: a door held, keys handed over
- L1: The intruder inside the building
- L4: A door held for a 'courier'; office keys handed to a stranger (62.5% without an awareness intervention)
- L6: Persuaded to let in
way out: Disclosure: credentials and information handed over
- L1: Credentials and live sessions reach the attacker; one-time codes relayed in real time
- L4: Credentials typed into a counterfeit login page, or read out on the phone
- L6: Persuaded to disclose
way out: Execution: an attachment opened, a found drive plugged in
- L1: Malware runs: backdoor access (RSA 2011); 45% of 297 dropped drives called home
- L4: A spreadsheet titled '2011 Recruitment Plan' opened; a found drive inserted
- L6: Persuaded to open or plug in
way out: Follow-on: the access used
- L1: Further accounts and systems: at least 11% of lateral phishers compromise another account
- L4: Access used: SecurID data used against Lockheed Martin; campaign emails leaked; others inside spear phished
- L7: The objective pursued: information gathering, fraud or system access
way out: Payment: money moved on a false instruction
- L1: Money leaves: nearly $47 million (Ubiquiti, 2015); $100 million (Google and Facebook, recouped); $3.05 billion in BEC losses reported to the FBI in 2025
- L4: Payment details changed or invoices paid on an impersonated instruction
- L6: Persuaded to pay
way out: Resistance: the lure reported
- L1: Reports feed newly found URLs to blocklists; takedown requested
- L4: Report button pressed: 18,476 reports in 15 months, about 10% of a campaign's reports within 5 minutes
- L6: Suspicion turned into a report