Phishing in 2026: What Abnormal AI and KnowBe4 Report

Phishing in 2026 is becoming more closely aligned with the ordinary systems, relationships, and workflows employees already trust. Two recent reports—Abnormal AI’s 2026 Attack Landscape Report and KnowBe4’s Phishing Threat Trends Report, Vol. 7—examine how attackers are making malicious activity resemble routine business communication.

The reports approach the subject from different directions. Abnormal AI concentrates on phishing, business email compromise, and vendor email compromise, comparing how tactics change by organization size, industry, job function, and geography. KnowBe4 covers a wider phishing environment that includes email, Microsoft Teams, calendar invitations, adversary-in-the-middle attacks, AI-supported campaigns, and attacker attribution.

Together, the reports are useful for security leaders, email security teams, SOC analysts, threat intelligence teams, identity specialists, and awareness practitioners. They show why phishing can no longer be treated only as a problem of suspicious messages entering an inbox.

Short answer

Both reports suggest that successful phishing increasingly depends on blending into trusted business activity rather than looking technically unusual. Attackers exploit familiar brands, legitimate platforms, internal identities, vendor relationships, collaboration tools, and routine processes. Abnormal AI provides the stronger analysis of how email attacks vary across organizational environments, while KnowBe4 provides broader coverage of AI, multi-channel phishing, AiTM techniques, and emerging delivery methods. The reports are complementary, but their statistics should not be compared directly because they use different datasets, time periods, classifications, and methodologies.

What the reports agree on

Attackers design lures around normal work

The clearest point of agreement is that phishing succeeds when the requested action appears consistent with the recipient’s role.

Abnormal AI finds that file-sharing phishing is concentrated in environments where document exchange is routine. It reports higher proportions of file-sharing lures among financial services, construction and engineering, finance and accounting, legal and compliance, and sales recipients. In these contexts, a SharePoint, Google Drive, Dropbox, or DocuSign notification does not automatically appear unusual.

KnowBe4 identifies the same underlying mechanism across a broader range of channels. Its examples include Workday impersonation in Microsoft Teams, HR-themed messages, calendar invitations, finance-related subject lines, and collaboration attacks that begin in email before moving to chat. The attack becomes credible because it fits an existing workplace process.

The common lesson is that organizations need to understand which communications are normal for each role. A document-sharing notification may deserve different scrutiny when it reaches a finance team than when it reaches an employee who rarely exchanges external files.

Trusted platforms are being used as attack infrastructure

Both reports describe attackers borrowing trust from established services.

Abnormal AI reports that brand impersonation appears in 12% of the phishing attacks in its sample. It also examines redirect chains, URL shorteners, legitimate brand assets, file-sharing services, and messages that combine mostly reputable infrastructure with one concealed malicious destination.

KnowBe4 reports that 22% of phishing attacks in its data are sent through a legitimate platform. It identifies PayPal, Google services, Microsoft, Zoom, and DocuSign among the platforms used to facilitate phishing. Its calendar-invite analysis also describes abuse of legitimate Google Calendar notifications and tracking URLs to support delivery and reputation laundering.

This means domain reputation and authentication results cannot independently establish whether a message is safe. A legitimate service can deliver malicious content, while a compromised account can send an authenticated message.

Credential theft remains central

Credential harvesting is a major concern in both reports.

Abnormal AI classifies phishing as attacks intended to steal credentials through malicious links, QR codes, or attachments. Its analysis covers redirect chains, shortened links, brand impersonation, file-sharing lures, and lateral phishing from compromised internal accounts.

KnowBe4 reports that 60.13% of identified attacks rely on malicious links and that 90% of malicious attachments include embedded credential-harvesting pages. It gives particular attention to adversary-in-the-middle phishing, in which a reverse proxy captures credentials and authenticated session cookies while the user interacts with a legitimate service.

The shared implication is that controls must address both credential entry and session theft. Multifactor authentication remains important, but KnowBe4’s analysis shows why some phishing techniques can capture an authenticated session after MFA has been completed.

Compromised accounts weaken traditional trust signals

Both reports emphasize attacks originating from real accounts.

Abnormal AI finds that lateral BEC becomes substantially more prominent as organizations grow. It accounts for 0.24% of BEC in small organizations but 23.2% in large enterprises in the report’s dataset. The report attributes this difference to the larger identity surface, greater number of trusted recipients, and increased return available from compromising an enterprise account.

KnowBe4 reports that 61.2% of phishing emails successfully bypassing security gateways come from compromised accounts. It also states that some of those accounts belong to trusted supply-chain organizations.

Once an attacker controls a legitimate account, sender authentication and domain reputation may support the malicious message rather than expose it. Detection therefore needs to consider changes in behavior, relationships, content, and requested actions.

Impersonation is shaped by context

Both reports treat impersonation as more than copying a senior executive’s name.

Abnormal AI identifies a substitution between VIP and employee impersonation. Among named-identity impersonation attacks, VIP impersonation falls from 43% at small organizations to 7% at large enterprises. In larger environments, peer or employee impersonation may be more credible than an unexpected message from the CEO.

KnowBe4 lists internal-team impersonation as its leading attack technique in the first quarter of 2026. Its Teams research also describes impersonation of IT, HR, finance, and executive personnel, sometimes reinforced through calls, multiple chat messages, or deepfake audio.

Security education should therefore cover requests from colleagues, departments, vendors, and collaboration accounts—not only obvious executive fraud.

Where the reports differ

Abnormal AI focuses on institutional context

The Abnormal AI report asks how an organization’s operating environment changes the attack methods it encounters.

Its analysis covers:

• Redirect and link-shortener use by organization size.
• File-sharing phishing by industry and job function.
• Brand impersonation and software-stack complexity.
• VIP, employee, departmental, and lateral impersonation.
• Vendor fraud by pretext, role, mailbox type, and geography.
• Differences between vendor impersonation and genuine vendor-account compromise.

This produces a detailed view of how attackers adapt to organizational structure. For example, the report finds that 81% of BEC attacks directed at shared mailboxes are vendor-related when standard and high-risk vendor email compromise are combined. Invoice fraud is especially compatible with accounts-payable and purchasing queues because those mailboxes process large volumes of external financial correspondence with limited relationship context.

The report is therefore particularly useful for organizations reviewing business processes, payment controls, shared inboxes, vendor verification, and identity behavior.

KnowBe4 focuses on expansion beyond the inbox

The KnowBe4 report treats phishing as a multi-channel and increasingly automated activity.

Its main areas include:

• Behavioral attribution of threat actors.
• Microsoft Teams phishing.
• Cross-channel attacks involving email and collaboration tools.
• AiTM phishing and reverse proxies.
• Phishing-as-a-Service toolkits.
• AI-assisted reconnaissance and content generation.
• Audio and deepfake payloads.
• Calendar invitation phishing.
• Indirect prompt injection against AI systems.

KnowBe4 reports a 41% increase in Teams-based attacks during the six months from October 2025 through March 2026. It also reports that 17.38% of Teams attacks in its analysis are multi-channel, beginning in email before moving to Teams.

This makes the report more useful for teams assessing collaboration security, session protection, AI-enabled social engineering, and attack paths that cross multiple applications.

The treatment of AI is substantially different

AI is not a central analytical category in the Abnormal AI report despite the source organization’s focus on AI-based detection. Its report primarily explains how attackers tailor phishing, BEC, and vendor fraud to operational context.

KnowBe4 presents AI as a major driver of phishing development. It reports that 85.8% of phishing attacks observed during the previous six months were AI-driven and estimates that AI-supported attacks are seven times more efficient than attacks relying on manual reconnaissance. It connects AI to personalization, polymorphic content, malware variation, voice cloning, larger attachments, and the emerging possibility of prompt-injection attacks against workplace agents.

These figures are specific to KnowBe4’s data and definitions. The provided material does not offer enough methodological detail to compare its AI classification directly with Abnormal AI’s dataset.

Vendor fraud is much more developed in the Abnormal AI report

Abnormal AI dedicates a substantial part of its report to vendor email compromise, which represents approximately 61% of BEC in its analysis.

It separates four principal pretexts:

• Invoice inquiry.
• Billing account update.
• Payment inquiry.
• Request for quote.

The report argues that attackers choose between impersonation and account compromise according to the credibility required. Only 0.95% of invoice-inquiry campaigns involve vendor-account compromise, compared with 26.5% of billing-account-update campaigns. A request to change banking details is more likely to receive scrutiny, making access to the genuine vendor account more valuable.

KnowBe4 covers BEC, compromised accounts, supply-chain trust, finance targeting, and impersonation, but it does not provide an equivalent breakdown of vendor-fraud pretexts in the supplied material.

Attacker attribution is more prominent in the KnowBe4 report

KnowBe4 profiles selected threat actors using behavioral consistency, campaign timing, volume, origins, attack types, and targeted industries. It contrasts short, high-volume activity with campaigns deliberately extended over several days to avoid volumetric detection.

Abnormal AI does not organize its findings around named threat groups. Its primary unit of analysis is the attack and the institutional environment in which it appears.

Threat intelligence teams seeking actor-level profiles may therefore find KnowBe4 more relevant. Teams redesigning email and payment controls may get more direct operational value from Abnormal AI.

How the methodologies affect the comparison

The reports should be read as complementary sources rather than competing measurements of a single population.

Abnormal AI provides a defined methodology. Its primary dataset contains 796,505 messages randomly sampled from 159.4 million attacks observed between July 1 and December 31, 2025. The sample covers 4,669 customer accounts across 43 countries and 21 industry categories. A separate census of more than 18,500 high-risk vendor-email-compromise campaigns supports its detailed VEC analysis.

The report also warns that its customer population is not representative of all organizations. Industry, geographic, organizational, and job-function categories are used to compare attack characteristics within segments, not to rank which groups receive the greatest absolute attack volume.

KnowBe4 states that, unless otherwise noted, its statistics come from its Collaboration Security products. Its report combines first-quarter 2026 data, year-over-year comparisons, six-month trends, expert commentary, threat-actor research, and individual technical investigations. The supplied material does not include a comparably detailed sampling appendix.

As a result, apparent differences—such as the proportion of attacks classified as BEC or the frequency of compromised-account delivery—may reflect different products, customer populations, classification rules, observation periods, and analytical units.

Key Cyntari themes related to this topic

Identity, Phishing & Access Security

This is the primary theme for both reports because they focus on phishing, credential theft, account takeover, impersonation, MFA bypass, and abuse of authenticated identities.

Cybercrime, Malware & Attack Techniques

Both reports examine attacker techniques including redirect chains, obfuscation, malicious attachments, reverse proxies, lateral attacks, social engineering, and multi-stage delivery.

Threat Actors, Geopolitics & Intelligence

KnowBe4 profiles named threat actors and campaign behavior, while Abnormal AI provides intelligence on how attack characteristics change by geography, industry, role, and organization size.

AI Security & AI Governance

KnowBe4 covers AI-assisted reconnaissance, polymorphic phishing, deepfakes, malicious agents, and indirect prompt injection. The theme is less central to Abnormal AI’s findings but remains relevant to the broader comparison.

Cyber Workforce, Awareness & Collaboration

Both reports show how attackers exploit employee routines, authority relationships, workload, trust, and communication norms. KnowBe4 also examines Microsoft Teams and calendar-based attacks directly.

SOC, Detection & Response

The reports discuss the limits of static indicators, domain reputation, secure email gateways, authentication checks, and signature-based controls when attacks use trusted infrastructure or legitimate accounts.

Software & Supply Chain Security

Abnormal AI’s vendor-email-compromise findings and KnowBe4’s discussion of compromised supply-chain accounts both show how trusted external relationships can become attack paths.

What this means for security teams

Map controls to business workflows

Security teams should identify workflows in which unusual requests can hide inside normal activity. Priority areas may include accounts payable, payroll, procurement, document exchange, legal transactions, shared mailboxes, IT support, and executive approvals.

Controls should reflect the risk of the action. A request to view a shared document does not require the same verification process as a request to change vendor banking information.

Extend phishing defenses beyond email

KnowBe4’s findings support treating Teams, calendars, calls, file-sharing services, and other collaboration systems as part of the phishing attack surface.

Employees should not assume that a request is legitimate because it appears in two applications. Cross-channel repetition can be an attacker’s method for manufacturing credibility.

Treat authentication as evidence, not proof

Passing SPF, DKIM, or DMARC does not mean a message is benign. Legitimate platforms, compromised business accounts, and attacker-controlled domains with correctly configured authentication can all deliver malicious communications.

Security tools need behavioral and contextual signals, including whether the sender normally contacts the recipient, whether the request fits the relationship, and whether payment or access details have changed.

Protect sessions as well as passwords

AiTM phishing demonstrates that account protection cannot stop at password security and basic MFA deployment.

Security teams should consider phishing-resistant authentication, session monitoring, conditional access, rapid token revocation, device context, and detection of unusual post-authentication behavior.

Build verification into financial processes

Abnormal AI’s findings make vendor and payment workflows a central defensive priority.

Banking-detail changes, invoice exceptions, new vendor requests, and unusual payment instructions should be confirmed through a trusted channel that does not rely on contact information supplied in the message itself. Shared finance inboxes require especially clear ownership and escalation procedures.

Adapt awareness training to roles

Generic phishing simulations may not reflect the lures employees are most likely to encounter.

Finance staff may need scenarios involving invoices and payment changes. Legal teams may need vendor-routing and document-sharing cases. IT teams may need helpdesk impersonation and MFA-reset scenarios. Sales teams may need fraudulent RFQs, while collaboration-heavy roles may need Teams and calendar exercises.

Recommended reports to read

For email, BEC, and vendor-fraud analysis

Abnormal AI — 2026 Attack Landscape Report — 2026 is the more relevant report for readers investigating how phishing and BEC vary by organization size, industry, role, mailbox type, and geography. Its vendor-fraud analysis is particularly useful for finance, procurement, fraud, risk, and email-security teams.

For AI, AiTM, and multi-channel phishing

KnowBe4 — Phishing Threat Trends Report — Vol. 7, April 2026 is the stronger choice for readers researching Microsoft Teams attacks, calendar phishing, reverse proxies, phishing-as-a-service, AI-assisted campaigns, deepfakes, and behavioral threat-actor attribution.

For a combined defensive review

Security leaders designing a wider phishing program should read both. Abnormal AI helps identify where malicious requests blend into business operations, while KnowBe4 helps identify how those attacks are spreading across channels and adopting new technical capabilities.