Abnormal AI leadership changes reflect rising AI-generated email threats

At a glance

What this is: Abnormal AI says AI-generated phishing and impersonation campaigns are increasingly bypassing secure email gateways, which it links to a 97% rise in advanced attacks over six months.

Why it matters: IAM and security teams need to treat email identity, user behaviour, and account compromise as a connected control problem because legacy detection loses precision when trusted communications can be convincingly imitated.

By the numbers:

• Advanced attacks bypassing Secure Email Gateways rose 97% over six months, according to Abnormal AI.

Context

Email security is no longer just a content-filtering problem. When attackers can generate messages that look operationally legitimate, the real control gap moves from signature-style detection to trust in sender behaviour, business context, and account authenticity.

In this article, Abnormal AI ties that shift to a 97% rise in advanced attacks bypassing secure email gateways over six months. The governance question for identity teams is how to distinguish legitimate communication from machine-generated impersonation without adding more friction to users or more exception paths to operations.

The leadership changes are the trigger, but the underlying issue is broader: organisations need controls that can evaluate communication behaviour, not just message format. That has direct implications for human IAM, account compromise response, and the way security platforms model trusted interactions across cloud applications.

Key questions

Q: Why do generative AI phishing campaigns bypass traditional email controls?

A: Because traditional controls often depend on patterns that attackers can now imitate or vary cheaply at scale. Generative AI can produce highly plausible wording, but it cannot as easily hide anomalous sending behaviour, abnormal reply chains, or unusual request patterns. Those behavioural signals are where defenders need more emphasis.

Q: How should security teams validate identity in AI-assisted email workflows to reduce impersonation risk?

A: Security teams should treat AI email assistants as untrusted intermediaries and require identity validation before any summary or action is used. The safest pattern is to verify the sender through authenticated headers, trusted account context, and a separate approval channel for high-risk requests. Do not rely on display names, because they can be spoofed and may be presented as trustworthy by the model.

Q: What breaks when email security relies mainly on static filters?

A: Static filters assume malicious messages can be identified from known patterns, but modern campaigns change structure, timing, and sender behaviour to avoid those rules. That leaves organisations exposed to convincing lures that trigger user action even when the message itself does not look obviously malicious. The failure is not just detection gaps, but an inability to follow the evolving campaign.

Q: What should organisations measure to know if email controls are actually working?

A: Organisations should measure detection fidelity, containment speed, and whether suspicious messages lead to fewer successful impersonation or credential theft events. A control is only effective if it changes attacker behaviour in production, not if it merely generates alerts or passes a policy review.

Technical breakdown

Why secure email gateways miss AI-generated impersonation

Secure email gateways were built to inspect static indicators such as sender reputation, domain patterns, payload signatures, and known malicious links. AI-generated phishing weakens those assumptions by producing fluent, context-aware messages that borrow the tone, timing, and vocabulary of real business exchanges. The result is not only better-looking lures, but lures that fit the expected behavioural shape of the organisation. When content alone becomes unreliable, the detection model has to move up a layer toward identity, intent, and communication pattern analysis.

Practical implication: reduce reliance on message-content rules alone and test whether your email controls can score behavioural anomalies.

Behavioral AI as a trust model, not just a detection layer

Behavioral AI in this context means correlating user, sender, device, and message behaviour to decide whether a communication fits a normal pattern. That is different from simple spam filtering because the control is trying to model legitimacy, not just block obvious abuse. For identity practitioners, this matters because email is often the first hop in account takeover, invoice fraud, and internal impersonation. The control challenge is therefore identity-adjacent: the platform must assess whether the apparent sender and the interaction path are consistent with the organisation's trusted communication graph.

Practical implication: align email security telemetry with identity and account-risk signals so suspicious communication can be evaluated in context.

Why connected applications expand the attack surface

The article notes deployment across Microsoft 365, Google Workspace, Slack, Workday, ServiceNow, and Zoom, which reflects the modern reality that email is only one node in a broader collaboration fabric. Attackers exploit that fabric by moving from email into chat, ticketing, HR, and workflow systems where trust often travels with the account. Once a message is accepted as plausible in one channel, it can be reinforced by activity in another. That makes governance harder because assurance now depends on how identity behaves across multiple cloud systems, not just inside the inbox.

Practical implication: inventory the downstream applications where email-born trust can be extended and monitor those paths for impersonation follow-through.

Threat narrative

Attacker objective: The attacker aims to convert believable machine-generated communication into trust, then use that trust to steal credentials, redirect payments, or expand into related cloud applications.

1. Entry begins with AI-generated email that mimics trusted internal or business communications closely enough to bypass secure email gateways.
2. The attacker uses behavioural realism, not obvious malware, to persuade a recipient to engage, trust the sender, or move into a second channel.
3. Impact follows when the message leads to account compromise, fraud, or further impersonation across connected applications and workflows.

Breaches seen in the wild

• McKinsey AI platform breach: McKinsey AI platform hack exposed 46M chats and sensitive data.
• OmniGPT breach claim 2025: A hacker claims to have leaked 34 million OmniGPT AI chat messages holding users' API keys and credentials; OmniGPT has not confirmed it.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 150+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Email trust is becoming a behavioural identity problem, not a filtering problem. When AI-generated threats can imitate the cadence and tone of legitimate business traffic, the old assumption that malicious mail looks obviously different stops holding. That changes the control objective from blocking bad content to assessing whether a message behaves like a trusted interaction. For identity programmes, this is a shift in where trust is decided, and it should be treated as such.

Human users are now the weak verification layer in a machine-assisted messaging environment. The article shows why training alone cannot carry the burden when messages are written to look operationally routine. Security teams have to assume that the first judgment point may be a busy employee, not a security control, which raises the value of contextual detection and account-level risk correlation. The implication is that human IAM and email governance need to be designed together, not separately.

Behavioral AI is emerging as a compensating control for the collapse of content-based email assurance. The most important issue is not that attackers use AI, but that they use it to erase the cues defenders relied on for years. That makes the problem an assurance problem: the organisation must decide which communication patterns are trustworthy enough to act on. Practitioners should treat behavioural analysis as part of trust establishment, not as a cosmetic enhancement to spam filtering.

Identity governance now extends into communication legitimacy across SaaS workflows. The article highlights a broader reality for enterprise security architecture: identity is no longer confined to login events and access requests. If a trusted message can seed activity in email, chat, and business applications, then governance has to follow the interaction chain. That means tighter linkage between account risk, anomalous messaging, and downstream application access.

Named concept: trust-boundary erosion in email. AI-generated communications collapse the boundary between legitimate business correspondence and adversarial impersonation. Once that boundary erodes, the defender can no longer rely on visual or linguistic cues to separate trust from threat. Practitioners should reframe email security around trust verification rather than message inspection alone.

What this signals

Trust-boundary erosion in email: defenders should expect the inbox to become a weaker source of truth as AI-generated messages blend into ordinary business workflows. That forces security teams to treat communication legitimacy as a policy and identity question, not only a mail-security question.

When email, chat, and workflow tools share the same trust assumptions, an impersonation attempt can move laterally without ever looking like a classic phishing chain. Security architecture has to follow the interaction path, not just the message.

Behavioral detection becomes the practical response when visual cues and language cues stop separating legitimate communication from adversarial content. Practitioners should watch for controls that can score interaction context across the enterprise, not just scan the inbox.

For practitioners

• Reassess email trust assumptions Map where your current secure email gateway decisions still assume that malicious messages look obviously malicious. Identify the cases where tone, timing, and business context now decide whether a message is accepted.
• Correlate email with identity risk Tie suspicious messaging signals to account compromise indicators, recent login anomalies, and unusual consent or privilege changes so the email event is not evaluated in isolation.
• Extend monitoring beyond the inbox Review which collaboration systems can continue an email-born attack path, including chat, HR, ITSM, and conferencing tools, and make sure they inherit the same trust scrutiny.
• Test behavioural detection gaps Run controlled phishing simulations that mimic legitimate internal communication patterns and measure whether your detection stack flags behavioural anomalies or only obvious indicators.

Key takeaways

• AI-generated email threats are eroding the reliability of content-based detection, which means communication trust has become an identity and behaviour problem.
• Abnormal AI reports a 97% increase in advanced attacks bypassing secure email gateways over six months, showing that the gap is already operational rather than theoretical.
• Security teams should connect email analysis to account risk, collaboration-system monitoring, and behavioural signals so trust is assessed before users act.

Key terms

• Behaviour-based email security: A security approach that judges email risk by how messages and accounts behave over time, not only by content or sender reputation. It looks for unusual reply patterns, impersonation signals, and identity-linked anomalies that traditional perimeter filters often miss.
• Trust-Boundary Erosion: The gradual loss of clear separation between legitimate interaction and adversarial impersonation. When attackers can copy tone, timing, and business context convincingly, defenders can no longer rely on visual or linguistic cues alone. The practical problem becomes deciding where trust should be established and verified.
• Behavioral AI: Behavioral AI is an analytics approach that looks for meaningful deviations in activity patterns rather than relying only on static indicators or signatures. In identity and security operations, it is used to identify suspicious sequences, unusual timing, and context shifts that suggest an attacker is adapting faster than conventional controls.
• Runtime Legitimacy: Runtime legitimacy is the question of whether the current actor presenting a valid credential still deserves access right now. It goes beyond token validity and focuses on process state, connection context, and whether the original trust assumption still holds.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are responsible for identity security strategy or NHI governance in your organisation, it is worth exploring.