Start free trial
Trending New Genres Authors
EnglishEnglish
EspañolSpanish
简体中文Chinese
繁體中文Chinese (Traditional)
FrançaisFrench
DeutschGerman
日本語Japanese
PortuguêsPortuguese
ItalianoItalian
한국어Korean
РусскийRussian
NederlandsDutch
العربيةArabic
PolskiPolish
हिन्दीHindi
Tiếng ViệtVietnamese
SvenskaSwedish
ΕλληνικάGreek
TürkçeTurkish
ไทยThai
ČeštinaCzech
RomânăRomanian
MagyarHungarian
УкраїнськаUkrainian
IndonesiaIndonesian
DanskDanish
SuomiFinnish
БългарскиBulgarian
עבריתHebrew
NorskNorwegian
HrvatskiCroatian
CatalàCatalan
SlovenčinaSlovak
LietuviųLithuanian
SlovenščinaSlovenian
СрпскиSerbian
EestiEstonian
LatviešuLatvian
فارسیPersian
മലയാളംMalayalam
தமிழ்Tamil
اردوUrdu
Searching...
SoBrief
Threat Modeling

Threat Modeling

Designing for Security
Find design-stage security flaws, then turn each risk into work the team can verify.
by Adam Shostack 2014 624 pages
4.07
286 ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Threat model early, while architecture is still cheap to change

Ask four practical questions. Threat modeling uses simplified models to find security problems before they become expensive code or deployment changes. The process asks: What are you building? What could go wrong? What will you do about it? Did the analysis hold up? A three-part web application, for example, can be sketched as a browser, web server, and database before implementation decisions have hardened. A design flaw found on that sketch is easier to address than the same flaw after customers depend on the system.

Treat it as practice. The method is a skill, not a document to produce once. Start with a real system, work through the questions, and repeat. The goal is not to imagine every possible disaster. It is to find actionable problems early, improve shared understanding, and make deliberate choices about security requirements.

Analysis

The timing argument aligns with a general principle in engineering economics: early decisions constrain later options, while revisions become more expensive as dependencies accumulate. The method also acts as a coordination tool. A diagram discussed before implementation can reveal that developers, testers, and operators hold different assumptions about the same component. Still, early analysis is not automatically accurate. Plans often omit the messy realities of deployment, so the model must be checked against the system as built, not treated as a prediction that cannot be revised.

Draw where trust changes, then trace every data path across it

Map the system's boundaries. A data flow diagram shows processes, data stores, outside participants, and the information moving among them. Add a trust boundary wherever control, identity, or privilege changes, such as between a customer browser and a company server, or between two organizations. These boundaries highlight likely attack points, but threats can also arise wherever attacker-controlled data is parsed or reused.

Keep diagrams useful. Show security-relevant decisions, data sources, destinations, and processes that move data between stores. If a flow sometimes uses encryption and sometimes does not, draw both cases. Split complex areas into readable subdiagrams rather than creating an unreadable wall of detail. A model is good when people can use it to explain the system and identify what might fail, not when it captures every implementation detail.

Analysis

A shared system diagram is more than a technical picture. It is a boundary object: a concrete artifact that lets specialists with different vocabularies discuss the same thing. This is useful in complex organizations, where misunderstandings often hide between teams rather than inside a single component. The emphasis on data movement also reflects how security failures cross layers: data safe in storage can become dangerous when passed to a parser or another service. The caveat is that diagrams can create false confidence if their omissions are invisible, so assumptions and scope need to remain explicit.

Use STRIDE to prompt discovery, not to force every threat into a box

Use a threat mnemonic. STRIDE prompts analysts to consider impersonation, unauthorized changes, denial or dispute of actions, exposure of information, service disruption, and users gaining powers they should not have. These are reminders of security properties that can fail, not a perfect classification system. One event may fit several prompts. Record the problem and its affected diagram element rather than wasting time arguing over its single correct label.

Match the tool to the team. Beginners can walk through each diagram element with STRIDE or use Elevation of Privilege, a card game that gives participants concrete threat prompts and points for identifying actionable issues. Attack trees map alternative paths to an attacker goal, while libraries such as CAPEC and OWASP provide more detailed patterns. These aids improve coverage, but no list proves that all relevant threats have been found.

Analysis

Structured prompts address a familiar problem in group reasoning: open-ended brainstorming favors confident specialists and can wander into spectacular but irrelevant scenarios. A mnemonic or card deck gives participants a common starting point and makes participation less dependent on security credentials. Yet every taxonomy shapes attention. A team can become skilled at finding the items its checklist names while overlooking a new interaction. The best use of structure is therefore generative: use it to start inquiry, then preserve space for observations that do not fit the categories.

Convert each credible threat into a decision, owner, bug, and test

Do not stop at discovery. Each threat needs a disposition: reduce its likelihood or impact, remove the feature that creates it, transfer the risk to another party, or explicitly accept it. For example, hiding an administrative webpage behind an obscure address does not remove the risk. Removing that web interface may eliminate that particular attack path, though the replacement access method still needs analysis.

Make the work trackable. File threats as bugs or tasks in the system the team already uses. Record who owns the fix, what was decided, and how success will be tested. A threat model for a database is incomplete if it lists exposed logs but never checks their permissions or tests whether they can be altered. Before delivery, verify that the diagram matches reality, every threat has a disposition, and planned tests have passed.

Analysis

This is a strong example of connecting analysis to an existing organizational control loop. A risk register that nobody uses is easy to ignore; a bug enters prioritization, staffing, testing, and release decisions. That does not guarantee a fix, but it makes deferral visible and accountable. The method also distinguishes uncertainty from acceptance: a documented choice can be revisited when conditions change, while an issue quietly dropped from discussion cannot. A useful refinement is to make the risk owner someone with authority over the affected system, not merely the person who recorded the issue.

Defend in layers, but cover ordinary attack paths before exotic ones

Prefer established defenses. Use platform features and well-tested techniques before inventing custom security mechanisms. Permissions can protect files, encryption can protect network traffic, and prepared database statements help keep user input from becoming executable commands. Validate data for its intended use and reject unexpected forms rather than relying on a broad cleanup routine. Isolation alone is fragile: even a network intended to be disconnected can acquire connections, devices, or malware over time.

Think in moves. A barrier can invite an attacker to try another route. Reinforced glass may address a broken-window attack, while an alarm addresses the next attempt to enter; the alarm itself then needs protection. Consider attacks against mitigations, but first cover the basic attack paths across the whole system. Cloud deployments add provider insiders, neighboring tenants, legal exposure, and possible limits on forensic evidence, so those boundaries belong in the analysis too.

Analysis

The chess-like view of defense is valuable because security controls alter an attacker's incentives rather than ending the contest. It connects threat modeling to resilience engineering, where systems are designed to contain failures and recover, not merely to prevent a single initiating event. The insistence on breadth before depth is also a safeguard against fascination with technically impressive edge cases. Still, standard controls are not automatically safe: a badly configured permission system or poorly managed encryption key can create a new failure. Reuse mature mechanisms, but test their actual deployment and operating assumptions.

Collect less personal data, because policy cannot erase what you retain

Model privacy as well as security. Privacy problems include surveillance, linking records, identifying people, using information for a new purpose, and excluding someone from a decision or service. The nymity slider describes how much identity information a transaction reveals. Greater linkability lets records from separate databases be combined into a richer profile. Research cited in the book found that birth date, gender, and ZIP code could uniquely identify 87 percent of the U.S. population.

Ask about social context. Contextual integrity examines whether information flows match the expectations of a setting, including who shares information, what it contains, and under what conditions. A hospital record used for treatment may become a privacy violation when repurposed for an unrelated decision. Privacy impact assessments and LINDDUN can help teams identify harms systematically. Data minimization is especially powerful: information that was never collected cannot be breached or repurposed.

Analysis

Privacy analysis broadens security's usual focus from unauthorized access to questions of legitimate use. A database may be perfectly protected from intruders yet still damage people through unexpected sharing or consequential inference. This connects technical design to social norms and institutional power: the same data flow can be appropriate in one relationship and intrusive in another. Minimization is a practical design principle, but it has real tradeoffs when information supports useful services or fraud prevention. Those tradeoffs should be discussed with affected people and made explicit, rather than hidden behind a broad consent screen.

Design accounts for changing people, not permanent names or secret facts

Separate identity from access. Authentication checks whether a claimant is associated with an account; authorization determines what that account may do. Accounts have life cycles: people join, change contact details, share access, leave, or die. A departing employee may retain access through a forwarding rule or recovery channel even after a password reset, so termination needs to address associated data and services.

Design recovery as a major attack path. Secret questions often rely on facts that relatives, social networks, or data brokers can discover. A documented account takeover chain showed how public details and weak recovery steps at several companies could combine to compromise email and other services. Biometrics also have limits: templates may be reconstructed, and unlike passwords, fingerprints cannot simply be replaced. Avoid treating government numbers or real-name policies as reliable proof of identity. Recovery should restore account access without exposing stored passwords.

Analysis

Account systems are socio-technical models of people, and their errors can outlast the original transaction. A mislinked record or recycled username may affect employment, finance, or reputation long after a technical incident is closed. The book's warning against overreliance on fixed identifiers is especially important as services combine data from multiple sources. However, less data is not always enough to resolve disputed identity or shared-account needs. Good design should support correction, delegation, and account closure, while making clear who has authority and how that authority can be challenged.

Make secure behavior the default, not a test of user vigilance

Treat people as part of the system. A ceremony is a protocol that includes human participants, their actions, and the information shown to them. People habituate to repeated warnings, work toward their main goal, and make quick judgments with whatever details are visible. The book describes a compliance budget: people have limited time and attention for security demands, so repeated friction can lead them to bypass controls.

Design for the real situation. Minimize decisions, avoid urgency, and provide an easy route back to safety. A protected document can open in a restricted mode that still allows reading, instead of demanding a risky yes-or-no choice immediately. Warnings should identify who is asking, explain the risk, offer realistic steps, and be tested. If a bank trains customers to distrust strange links but sends its own messages through unfamiliar domains, it creates scamicry, behavior that resembles a scam and weakens useful cues.

Analysis

This section rejects the convenient fiction that security failures are simply user mistakes. Human-factors research shows how interface habits, divided attention, and poor feedback shape behavior. Moving security decisions into architecture can reduce cognitive burden and improve outcomes for people who are busy, stressed, or unfamiliar with the technology. Yet automation also concentrates power in system designers: a secure default can become an unwanted constraint if users cannot understand or override it. The strongest design combines a safe default with an intelligible explanation and a meaningful path to appeal.

Use standard cryptography and treat keys, endpoints, and upgrades as the real work

Do not invent cryptography. Encryption, signatures, hashes, and randomness serve different purposes; confusing them can leave data exposed or modifiable. Use established algorithms and libraries rather than a home-built cipher. A password hash should be salted and deliberately slow to make mass guessing more costly. Encryption alone does not guarantee integrity, and a secure channel does not protect a message after it leaves that channel.

Protect the whole arrangement. Cryptography depends on sound key generation, storage, revocation, and identity checks. A certificate can bind a public key to a name, but the software must still verify that the name is the intended destination. Kerckhoffs's principle assumes the system design may become public, so security should depend on protected keys rather than secret code. Protocol flexibility can increase complexity and allow an attacker to force weaker options; upgrades should be designed and tested from the start.

Analysis

Cryptography demonstrates a broader engineering lesson: a powerful component does not make the surrounding system sound. The strength of an algorithm can be irrelevant if a key is embedded in the application, a user accepts the wrong certificate, or an old protocol remains available. This is why secure systems are better understood as chains of assumptions than as collections of impressive primitives. There is also a usability dimension: mechanisms that require constant human judgment invite routine errors. Standard libraries reduce avoidable design risk, but integration and key governance still require careful ownership.

Make threat modeling a repeatable team practice, not a report that gets shelved

Fit the work to the organization. Threat modeling can begin with a broad system view, then proceed across boundaries and components. Bring together people who understand implementation, testing, operations, and business goals. Use diagrams and bugs as shared artifacts, assign responsibilities, and attach the work to planning and release routines. A short, well-scoped exercise that finds a fixable problem can build more support than a sweeping process nobody can sustain.

Learn from outcomes. Revisit the model when architecture changes and check it before delivery. If an incident exposes a missed threat, ask whether the diagram omitted a flow, the discovery method missed a class, or the team accepted the risk without recording it. The book also argues for flow: clear goals, useful feedback, and challenges suited to participants' skills. Measure carefully, since a score or threat count can reward paperwork rather than better security.

Analysis

The organizational argument is as important as the technical one. A practice survives when its outputs are useful to the people who must act on them, and when its workload fits the development process. This resembles high-reliability work in aviation and medicine, where checklists and review rituals help teams catch omissions, but only when they are integrated into real decisions. Metrics remain hazardous: counting diagrams or threats can encourage volume over insight. A stronger signal is whether analysis changes a design, produces a tested mitigation, or clarifies a risk that an accountable decision-maker consciously accepts.

Analysis

The book's most important move is to turn security from a specialist's hunt for clever attacks into a shared engineering activity. Its four questions create an intelligible path from system representation to threat discovery, response, and validation. The approach is deliberately pragmatic: it favors software diagrams because builders understand them, and bugs because development organizations already know how to manage them. That practical choice also carries a theory of change. Security improves when the people who make design decisions can see how data crosses boundaries, name a failure, and route the work into existing processes.

The book's strongest insight is that a threat model is not just a list of adversaries or valuable assets. It is a working abstraction that coordinates different forms of expertise. A diagram can surface disagreement; a threat prompt can make implicit assumptions discussable; a bug can turn analysis into an obligation. In this sense, threat modeling is both technical analysis and organizational infrastructure. It reduces the distance between security knowledge and the people who can implement it.

That same pragmatism sets limits. STRIDE, attack trees, and libraries improve consistency but cannot establish completeness. Diagram quality depends on what the team knows to include, and risk prioritization remains partly political: organizations decide whose harm matters, which uncertainties deserve investment, and what is acceptable to transfer to customers. Human behavior and privacy are especially resistant to purely technical decomposition. They require contextual knowledge and attention to power, incentives, and expectations.

The durable lesson is therefore not to worship a particular mnemonic or tool. It is to build a revisable practice with explicit scope, diverse participants, traceable decisions, and tests that challenge assumptions. A useful threat model does not predict every attack. It helps a team notice consequential failure paths while it still has choices, and gives those choices an owner, rationale, and feedback loop.

Summary in 30 Seconds
Represent systems with data-flow diagrams that expose trust boundaries, then use structured prompts to find plausible failures. Assign each risk a decision, an owner, and a test, and revisit the model as the system changes. Reduce exposure by collecting less personal information: birthdate, gender, and ZIP code together can identify 87 percent of Americans. Use established cryptography, protect recovery paths, and make safer actions practical for users.
Contains spoilers

Last updated:

Report Issue

Review Summary

4.07 out of 5
Average of 286 ratings from Goodreads and Amazon.

Threat Modeling receives mostly positive reviews for its comprehensive coverage of threat modeling techniques, particularly focused on the STRIDE methodology. Readers appreciate the practical guidance, artifacts, and insights from real-world experience. The book is praised for addressing "blank page paralysis" and providing valuable information for both novices and experienced practitioners. Some criticisms include excessive detail in certain sections, loose terminology, and a heavy focus on Microsoft's perspective. Despite these issues, many reviewers consider it a valuable resource for security professionals and software developers.

Your rating:
4.51
111 ratings
Want to read the full book?

Glossary

Threat modeling

Structured analysis of risk

The use of simplified representations of a system and possible threats to identify security problems, choose responses, and check whether the analysis is sound. In this book, it is a practical activity for software teams and system operators, not just a document or an exercise reserved for security specialists.

Four-step framework

Model, find, address, validate

A structure for threat modeling organized around four questions: what is being built or deployed, what can go wrong, what should be done about those problems, and whether the analysis was adequate. Different diagrams and threat-finding methods can be combined within this structure.

trust boundary

A change in control or privilege

A point where entities with different control, identity, or privileges interact. Examples include a network connection between organizations, a transition between user accounts, or communication between processes with different rights. Threats often cluster at these boundaries, though they can also occur anywhere attacker-controlled data is processed.

STRIDE

Six prompts for security failures

A mnemonic for prompting analysis of common security failures: impersonation, unauthorized modification, denial or dispute of actions, exposure of information, interruption of service, and excessive privilege. The book treats it as a discovery aid, not a rigid classification scheme. A single attack can raise several prompts, and it is more important to record the issue than settle its label.

Elevation of Privilege

A threat-modeling card game

A collaborative game that uses cards to prompt players to connect concrete security threats to a diagram of a system. Players identify issues they consider actionable, helping newcomers participate without relying only on open-ended brainstorming. The game is designed to make threat discussion structured and approachable, not to prove that a model is complete.

attack tree

Paths toward an attacker goal

A branching representation of ways an attacker might achieve a goal or cause a harmful condition. Branches may indicate alternatives or combined requirements, so the creator must make the logic clear. Attack trees can help discover threats, organize known threats, or communicate an analysis, but their completeness and scope are difficult to judge.

ceremony

A protocol including human actions

An extension of protocol analysis that includes people, interfaces, physical actions, and communication between people as parts of the system. A ceremony analysis asks what information each participant receives, what decisions they must make, and how attacks can exploit missing, confusing, or misleading cues.

contextual integrity

Privacy shaped by information context

A privacy framework for judging whether information flows fit the norms of a social setting. Analysis considers the context, the people or organizations involved, the information being shared, and the conditions governing its transmission. A data flow expected in one setting can violate privacy when the participants, purpose, or sharing rules change.

LINDDUN

A privacy-threat discovery mnemonic

A structured approach to privacy threat modeling that prompts analysis of linking records, identifying people, denying or disputing actions, detecting activity, disclosing information, lack of user awareness, and failure to follow policy or consent. It can complement or accompany security threat discovery, though its specialized terminology may require training.

scamicry

Legitimate behavior resembling a scam

A legitimate organization's behavior that is difficult for ordinary people to distinguish from a scammer's behavior. Examples include requesting sensitive information through an unexpected channel or using unfamiliar tracking links. Scamicry weakens people's ability to recognize suspicious behavior and can train them to ignore otherwise useful security advice.

conditioned-safe ceremony

A routine that trains safe action

A user interaction designed so that repeated use encourages a safe response. Its rules should remain safe even when an attacker controls the situation, and the design should not depend on users reliably making untrained choices. Such routines can reduce decision burden, but must be tested to ensure the safe action is practical and hard to bypass.

compliance budget

Limited attention for security tasks

A model of the time and effort people are willing or able to spend on security demands amid their other responsibilities. Repeated or confusing requirements can exhaust this budget, leading people to skip steps or work around controls. Designers should reduce unnecessary security demands and make required actions feasible in the actual work context.

About the Author

Adam Shostack is a renowned expert in cybersecurity and threat modeling. He has extensive experience working at Microsoft, where he was a leader in the Security Development Lifecycle (SDL) team. Shostack is known for his contributions to improving software security practices and has been instrumental in developing and promoting threat modeling methodologies. He is a frequent speaker at security conferences and has authored multiple books on cybersecurity topics. Shostack's work has significantly influenced the field of threat modeling and secure software development. His approach emphasizes practical, actionable techniques that can be applied across various organizational contexts.

Download PDF

To save this Threat Modeling summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.30 MB     Pages: 17

Download EPUB

To read this Threat Modeling summary on your e-reader device or app, download the free EPUB. The .epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.
Download EPUB
File size: 1.34 MB     Pages: 22
Want to read the full book?
Follow
Listen
Home
Swipe
Library
Get App
Try Full Access for 3 Days
Listen, bookmark, and more
Compare Features Free Pro
📖 Read Summaries
Read unlimited summaries. Free users get 3 per month
🎧 Listen to Summaries
Listen to unlimited summaries in 40 languages
—
❤️ Unlimited Bookmarks
Free users are limited to 4
—
📜 Unlimited History
Free users are limited to 4
—
📥 Unlimited Downloads
Free users are limited to 1
—
Risk-Free Timeline
Today: Get Instant Access
Listen to full summaries of 26,000+ books. That's 12,000+ hours of audio!
Day 2: Trial Reminder
We'll send you a notification that your trial is ending soon.
Day 3: Your subscription begins
You'll be charged on Nov 24,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
1M+ readers
Trustpilot Rating
TrustPilot
4.6 Excellent
This site is a total game-changer. I've been flying through book summaries like never before. Highly, highly recommend.
— Dave G
Worth my money and time, and really well made. I've never seen this quality of summaries on other websites. Very helpful!
— Em
Highly recommended!! Fantastic service. Perfect for those that want a little more than a teaser but not all the intricate details of a full audio book.
— Greg M
Save 62%
Yearly
$119.88 $44.99/year/yr
$3.75/mo
Monthly
$9.99/mo
Start a 3-Day Free Trial
3 days free, then $44.99/year. Cancel anytime.
Unlock a world of fiction & nonfiction books
26,000+ books for the price of 2 books
Read any book in 10 minutes
Discover new books like Tinder
Request any book if it's not summarized
Read more books than anyone you know
#1 app for book lovers
Lifelike & immersive summaries
30-day money-back guarantee
Download summaries in EPUBs or PDFs
Cancel anytime in a few clicks

Why one summary?

Scanner
Find a barcode to scan

⭐
⭐
⭐
We have a special gift for you
Open
38% off
DISCOUNT FOR YOU
⭐
⭐
⭐
$79.99
$49.99/year
Only $4.16 per month
Continue
2 taps to start, super easy to cancel
Settings
General
Widget
Loading...