The FBI’s Internet Crime Complaint Center reported more than 22,000 complaints that involved AI in 2025, with nearly $900 million in associated losses — and recorded 24,768 business email compromise (BEC) complaints that year with about $3.05 billion in reported losses.
Generative AI and the loss of behavioral tells
Proofpoint told attendees at its Protect 2026 conference that generative AI has removed many of the shortcuts attackers once relied on — and with them the behavioral fingerprints defenders used to catch them. Where attackers formerly reused templates and infrastructure, leaving language mistakes, timing oddities, and repeatable artifacts, language models now produce fluent, localized text with matched signatures, fabricated conversation histories, and unique messages for each recipient.
- Language: fluent, localized text with no typos or translation artifacts.
- Context: replies that fit the thread, the recipient’s role, and the timing.
- Persona: matching writing style and fabricated histories.
- Variation: unique, per-recipient messages generated in seconds so there is no template to match.
The real estate campaign: thirty distinct approaches, one ask
Proofpoint presented a concrete example: a single actor contacted more than 30 real estate agents at the same company, each from a different account, each about a different property and in different words — and every thread closed with the same request: join a video meeting. One inspection enquiry sent at 2:58 a.m. became a branded meeting invitation 13 minutes later that claimed “immediate attention.” The join link used a genuine google.com redirect so reputation checks appeared clean, while a URL parameter handed the click to an unrelated site.
Because each sender was making first contact and every message was different, behavioral filters had no history or repeated content to match. Proofpoint’s point: the common signal was not how the messages looked but what they were trying to get recipients to do.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scramblePhishing-as-a-service, device-code attacks, and hostile messaging campaigns
Proofpoint also highlighted infrastructure-level changes. A phishing-as-a-service kit was shown that exposes a single flow behind multiple disguises — court portals, document-signing requests, identity checks, finance reports — each displaying a verification code and directing victims to a legitimate sign-in page to paste it. Because the attacker generated the code, pasting it authorizes the attacker’s session, a technique Proofpoint described as device-code phishing.
Separately, Proofpoint described hostile communications campaigns where hundreds or thousands of senders each write tailored messages to the same executives. Some include derogatory language or AI-generated images that mock the recipient; none carry a detonating link or credential harvest in isolation. Individually, any one of these messages could appear to be a single angry correspondent — but taken together they express disruptive intent that behavioral scoring of individual messages misses.
Proofpoint Nexus and intent-based detection
To meet this shift, Proofpoint announced a Nexus Intent-Based Detection Model in September as part of an Agentic Collaboration Security system that reads what a message is trying to achieve rather than relying only on how it looks. Proofpoint positions intent analysis as a third layer that complements traditional detection (rules, signatures, reputation) and modern detection (threat intelligence and behavioral baselines).
Proofpoint illustrated intent with a bank-detail change request where five details indicate malicious intent despite a clean-looking message:
- It is the supplier’s first banking change in years.
- The request does not match the invoice-confirmation thread it replies to.
- The deadline is tied to a specific payment run.
- The sender asks to keep the exchange off the thread and deal with one person.
- Verification is closed off in advance with a claim that the finance mailbox is “mid-migration.”
The Nexus approach also maps detection depth to mail-flow needs with three tiers: flash models that evaluate all mail in under 500 milliseconds; an extended tier handling fewer than 1 percent of messages in seconds to minutes; and a deep tier handling fewer than 0.01 percent in minutes with multi-stage language-model analysis, voice and video, and privileged-person profiling. Messages escalate until a tier can resolve them, and a malicious verdict can arrive at any tier.
How technologists, enterprises, and end users should watch and act
- Technologists and security teams: scale intent reasoning and agentic investigation to match machine-speed campaigns. Proofpoint’s Threat Investigation Agent is designed to run one investigation across email, identity, messaging, browser, and supplier activity and map a campaign’s blast radius.
- Enterprises and procurement leaders: pay attention to processes that accept first-time contacts — new-client intake, clinics taking new patients, banks opening accounts, insurers quoting policies — since Proofpoint demonstrated that AI can tailor believable first contacts across those front doors.
- End users and business owners: scrutinize requests that seek off-thread handling, urgent payment deadlines, or one-person verification steps. Proofpoint’s examples show these signals persist even when language and sender reputation are polished.
Proofpoint’s demonstrations and the FBI complaint totals point to a clear pivot: generative AI has stripped away many traditional behavioral cues, forcing detection toward reasoning about intent and campaign-level patterns. Proofpoint invites readers to learn more and see a demo at Proofpoint Nexus.
