Mitigation · OAK-M45 · user-behavioural
OAK-M45 — Inbound-Contact Refusal and Out-of-Band Callback
Description
A channel-level invariant that resolves an entire class of impersonation without requiring the user to evaluate the impersonator: treat inbound contact claiming to be support as the attack, terminate it, and re-establish the channel from a contact point held before the contact occurred.
The mitigation's leverage comes from moving the decision away from content and onto direction. Content-based defence — spotting a wrong domain, an odd phrasing, a bad accent, a spoofed number — is a contest the defender loses over time, because every content signal the defender learns is a signal the attacker can buy or fix. Direction is not improvable by the attacker: to run the attack at all they must initiate. A rule that refuses all inbound support contact therefore has no exploitable gradient, which is why it outperforms every "how to spot a scam call" checklist in the corpus.
The rule's second function is defeating the manufactured clock. Support-impersonation scripts universally assert that the account is being drained now and that hanging up will cost the victim their money. That urgency is not incidental — it exists specifically to prevent the callback, because the callback is the only step that reliably kills the attack. Recognising urgency-to-prevent-verification as the tell inverts the script's own mechanism: the harder the caller resists being called back, the more certain the diagnosis. Naming this explicitly is the load-bearing part of the mitigation, because the emergency is precisely what makes a normally-careful person skip the verification step.
How it applies
- OAK-T11.007.003 (Brand-Trust-Leveraged Active Phishing for Seed-Phrase Exfiltration): the technique is channel counterfeiting — the device is genuine and the communication is attacker-controlled. Refusing the inbound channel and reconnecting through the vendor's published support path defeats the campaign at its only load-bearing step, independent of how well the counterfeit channel is executed. Pairs with
OAK-M44, which holds as terminal defence if the channel rule fails. - OAK-T11.005.001 (Fake-CEX / Pig-Butchering Platform): acquisition for this technique is inbound by construction — an unsolicited message, a wrong-number opener, a dating-app match, a friendly stranger with an investment insight. The rule applies to the relationship, not just to support calls: an investment platform that arrives via inbound contact from someone who found you is the technique's own entry vector.
- OAK-T11.005.002 (Fake-Custodian / Fake-Asset-Manager Fraud): same acquisition shape, arriving as MLM recruitment or an "AI-trading" pitch. Refusing inbound-sourced investment platforms removes the reader from the funnel before the fattening phase begins, which is the only phase where refusal is cheap.
- OAK-T4.010 (Fake Security-Tool / Browser-Extension Phishing): a significant sub-flow is inbound "we noticed suspicious activity on your wallet — install our scanner" contact. The rule intercepts it at the channel;
OAK-M44intercepts it at the payload.
Limitations
The largest inbound-impersonation surface in the corpus has no Technique to map to. Voice/phone impersonation of a custodial venue's support to an end user — the Coinbase-support-impersonation class, ~$300M/yr, with named arrests — is currently mapped to OAK-T4.007, whose scope is engagement-weighted in-platform distribution mechanics (comment-pinning, reply-ranking) and does not cover a spoofed phone call. T11.007.003 is parented under hardware-wallet compromise and does not reach exchange support either. This mitigation therefore under-declares its real coverage, and will keep doing so until the gap is filled. See TAXONOMY-GAPS.md.
Does nothing against outbound-initiated compromise: a user who searches for support and lands on a paid ad for a fake support line (T4.008) has initiated the contact themselves and this rule never fires. That surface belongs to OAK-M47.
Does nothing once the relationship is established. The rule is an acquisition-phase control; a victim three months into a pig-butchering relationship is no longer receiving inbound contact from a stranger, they are talking to someone they believe they know. By the deposit phase, this mitigation has already failed and OAK-M46 is the remaining line.
The rule also carries a real false-positive cost that should be stated rather than minimised: legitimate institutions do make outbound calls, and a user applying this rule will hang up on some of them. That cost is correct to pay — the callback resolves the legitimate case at the price of a few minutes — but a mitigation that pretends it is free will not be followed.
Reference implementations
- Venue-side (partial, uneven): in-app-only support channels with no outbound voice; published verification that support never calls; caller-ID anti-spoofing at the carrier layer (STIR/SHAKEN in the US, patchy internationally and trivially bypassed via VoIP origination).
- Vendor-side: published "we will never call you" statements. The corpus indicates these do not reach the user at the moment of the call.
- No production tooling. This is a human refusal executed under engineered stress. The nearest structural control is venue-side elimination of the outbound-voice channel entirely, which removes the attacker's cover story.
Citations
- See
examples/2023-2026-coinbase-support-impersonation.md— the class anchor; ~$300M/yr; data-broker-sourced personal detail; named arrests (rare in the corpus). Currently mis-mapped to T4.007 — see Limitations. - See
examples/2024-08-genesis-creditor-social-engineering.md— the class's best-established anchor and the one this mitigation is actually specified against: two chained spoofed calls an hour apart (Google Support, then Gemini Support, each stage's extraction authenticating the next), $243M, attributionconfirmed, three named principals, one arrest. The callback rule breaks the chain at either stage. - See
examples/2026-01-hardware-wallet-social-engineering.md— $282M+, larger but weaker evidence: no public source establishes the impersonation channel, so whether this mitigation would have reached it is unknown. Cited for scale, not as a mechanism anchor. - See
examples/2024-01-kk-park-compound-takedown.md— the inbound acquisition script at industrial scale.