The problem

A breach usually starts with one ordinary-looking email

300 staffEmailCloud driveFinance systempays suppliersAttackerInvoice 2291 shared with you“Sign in to view the document.”Sent to 40 people in sales and financeA breach usually starts with one ordinary-looking email. Follow it all the way to the money.A company of 300 people uses cloud email, a shared drive and a finance system that pays suppliers. One morning 40 people in sales andfinance receive an email saying an invoice has been shared with them, with a link to sign in. Nothing about it is unusual: invoicesare shared like this every day. This case walks the attack link by link, and asks at each one: what would have stopped it here?

An invoice-sharing email with a sign-in link reaches 40 people in sales and finance.

The attack chain

An attack is a chain, and each link needs the one before it

Emailto 40 staff:“shared invoice”Fake sign-in pagea lookalikedomainPassword stolenattacker signs inas Dana in salesMailbox rulehides replies aboutinvoicesFake request“new bank details”$48,600 paidOf the email40 received, 3 clicked, 1 typed a passwordNine days laterthe real supplier asks why invoice 2291 has not been paid. That is how it is found.An attack is a chain of steps. Each step needs the one before it, so each one is a place to stop it.3 of 40 people click and one, Dana in sales, types a password on the fake page. The attacker signs in as Dana from another country, addsa rule that moves any reply mentioning invoices into a folder Dana never opens, then writes to finance from Dana’s real mailbox: thesupplier has new bank details. Finance pays $48,600. Nobody notices for 9 days.

3 click, 1 types a password. A mailbox rule hides replies, and finance pays $48,600 to new bank details.

The first defence

Training is a filter, not a wall

40 emailstrainingGot through3 clicked (7.5%), 1 typed a passwordTraining lowers the number. It never reaches zero, and the attacker needs only one.Training is a filter, not a wall: it lets a few through, and one is enough.These staff had yearly awareness training, and 37 of 40 did the right thing. 3 still clicked, which is 7.5%: a good result for aconvincing email. Training is worth doing, and reports from staff are how many attacks are spotted. But a defence that works onlywhen every person gets it right every time is not a defence on its own. The next layers must hold when a person slips.

Training brought it down to 3 clicks in 40. The attacker needs only one.

The cause

Each link held because one control was missing there

EmailFake sign-in pagePassword stolenMailbox ruleFake requestno MFA:a password was enoughno alert ona new mailbox ruleone person couldchange bank detailsTraining: workedfor 37 of 40Spam filter: missed ita new, clean domainEach step went through because one control was missing there, not because the attacker was brilliant.Email filtering and training did their part, as well as they ever do. After that, a password alone was enough to sign in from abroad, anew rule hiding mail raised no alert, and one person in finance could change a supplier’s bank details on the strength of an email.Three missing controls, each cheap. Any one of them would have broken the chain.

No MFA at sign-in, no alert on a new mail rule, and one person could change bank details.

The fix

Three controls, one per link; any one stops the chain

EmailFake sign-in pagePassword stolen✗Mailbox rule✗Fake request✗Passkey sign-inwill not workon a fake siteAlert on mail rulesthat hide orforward mailCall, then approvea known number,and a second personThree controls, one per link. Any one of them stops this attack; together they stop most of its cousins.A passkey or security key signs in only to the real site, so a password typed on a fake page is useless. An alert when anyone creates arule that hides or forwards mail would have fired within minutes. And a change of bank details is confirmed by calling the supplier ona number already on file, then approved by a second person. That is defence in depth: no single control has to be perfect.

A passkey at sign-in, an alert on new mail rules, and a phone check with two approvers for bank changes.

Respond and recover

Response runs in the same order; the alert decides when it starts

Detectrule alert after 4 minContainreset, revoke sessionsEradicatedelete the ruleRecovercall the bank, recallLessonsthe three controlsTime to detect, to scaleSupplier asks9 daysRule alert4 minutes, before any paymentResponse follows the same order every time; the alert decides whether it starts in minutes or days.Contain first: reset Dana’s password, revoke every signed-in session and token, and block the lookalike domain. Eradicate: delete the rule,check the other 2 who clicked and search every mailbox for similar rules. Recover: ask the bank to recall the payment at once, sincerecalls succeed far more often within hours than after days. Found after 9 days, the money was gone; found in 4 minutes, none would have left.

Detect, contain, eradicate, recover, learn. Detection took 9 days; an alert would take 4 minutes.

Defence in depth

Layers around what matters, so one failure is not the end

Awareness and reportingLesson 3Phishing-resistant sign-inLesson 7Least privilegeLesson 8Logs and alertsLesson 11Two approvers for paymentsLesson 1A practised responseLesson 12Defence in depth: layers around what matters, so that one failing layer is not the end.The attacker has to pass every ring to reach the money; the company needs only one to hold. The rings are not all technology: twoof them are a process and a habit. Each ring opens the lesson that teaches it. The rings guard the finance system because the attackerwanted the money, not the mailbox: start any design from what an attacker is after.

Awareness, sign-in, least privilege, alerts, two approvers, response. Each ring links to its lesson.

A real case

At Twitter in 2020, a phone call and a fake VPN page

Phone call“IT help desk:VPN problem”Fake VPN pagecredentials typedin real timeMFA promptsome staffapproved itInternal toolsreached adminaccess130 accounts45 tweeteda bitcoin scamWhat the report recommendedOur caseHardware security keys, not app MFAphishing-resistant MFASecond-person approval for sensitive actionstwo approvers for paymentsMonitoring for anomalies (SIEM)alert on mailbox rulesAt Twitter in July 2020, a phone call and a fake VPN page reached 130 of the world’s best-known accounts.Per the New York DFS report, attackers called employees posing as the IT help desk, sent them to a fake VPN login, entered what theytyped into the real site and some employees approved the MFA prompt. With internal tools they took over 130 accounts and collected about$118,000 in bitcoin in roughly a day. The report’s fixes are this case’s fixes: keys that cannot be phished, two people, and monitoring.

Callers posing as the IT help desk phished staff and their MFA approvals, then took over 130 accounts.

Lesson complete

You finished “One email, one breach”

Nicely done.

Open in a new tab