AI-Powered Phishing Is Evolving: Why Tenant-Specific, Document-Based Lures Are the New Social Engineering Threat

By BCyber Security Team | CISSP, CISM Last updated: 26 June 2026

Disclosure: This article references BCyber CyberSuite products and BCyber platform data because it is published by BCyber. The recommendations below reflect our implementation and assessment experience across client environments.

Phishing used to be noisy, generic, and easy to spot if you knew what to look for. Today, AI-powered phishing is becoming quieter, more targeted, and far more convincing because it is built to resemble normal work inside a real organisation. Instead of obvious bad grammar and spoofed logos, attackers are using tenant-specific clues, collaboration workflows, and document-based requests that feel routine.

That shift matters in Australia because cloud-first organisations increasingly rely on shared files, comments, approvals, and external invites as part of day-to-day operations. In regulated sectors such as financial services, healthcare, and government, the consequence is not just credential theft; it can become a control failure under APRA CPS 234, a maturity gap against the Essential Eight, or a privacy incident under the Privacy Act 1988 and Notifiable Data Breaches scheme.

From broad bait to tenant-aware deception

Traditional phishing campaigns were built for scale, not subtlety. They depended on mass distribution, generic lures, and the hope that a few recipients would click before thinking too hard. Modern attackers still want scale, but AI lets them customise their messages fast enough to make each lure feel local, relevant, and time-sensitive.

When we say tenant-specific phishing, we mean a lure built around the target organisation’s own digital environment. Attackers may reference real departments, names, vendors, document naming conventions, or the kind of workflow language employees use every day. A message might look like a shared file notification from Microsoft 365, a Google Workspace doc request, a Slack prompt, or a Salesforce-related update that seems to come from a familiar process.

This matters because users tend to trust what fits the context they already expect. If a message mirrors a real approval chain, a common file-sharing habit, or a known vendor relationship, it lowers the chance of immediate suspicion. Security tools can also struggle when the lure uses legitimate cloud services, compromised accounts, or content that does not contain an obviously malicious attachment.

The most dangerous phishing message is often the one that looks like work already in progress.

Why the Australian regulatory context matters

For Australian organisations, this is not just a user-awareness issue; it is a governance and resilience issue.

• APRA CPS 234 expects regulated entities to maintain information security capability commensurate with threats and asset criticality. If an AI-generated lure leads to account takeover, the organisation’s ability to detect, contain, and evidence control operation becomes the real test.
• The ACSC Essential Eight still matters because phishing remains a common entry point for malicious macro abuse, session theft, and privilege escalation. Even when attackers do not deliver malware directly, they often use the initial compromise to reach systems the Essential Eight is meant to harden.
• The Privacy Act 1988 and Notifiable Data Breaches scheme raise the stakes for healthcare providers, professional services firms, and any business handling sensitive information. A compromised inbox or cloud tenant can expose personal, financial, and health information faster than teams can manually contain it.

In practice, we see the same pattern across sectors. A financial services team receives a fake board-pack or procurement approval file that looks like it was shared by a known executive assistant. A healthcare provider sees a referral summary, care plan, or policy update sent through a collaboration link. A government-adjacent team is asked to review a briefing note, contract variation, or supplier statement in a document workflow that matches normal business processes.

The lure succeeds because it feels operational, not suspicious.

How AI turns reconnaissance into believable requests

AI gives attackers speed, variety, and polish. It can generate multiple versions of the same email in seconds, adjust tone for a specific role, and translate content naturally enough to avoid the awkward phrasing that once gave scams away. That means attackers can test subject lines, urgency levels, and document formats until one lands.

The other advantage is reconnaissance at scale. Attackers can use public websites, job ads, social media profiles, leaked documents, and file metadata to build a picture of who does what, which tools are used, and how decisions get made. AI can then summarise that information into a useful trust map showing which names, documents, and relationships are most likely to trigger action.

This is especially dangerous in environments with business email compromise (BEC) exposure, OAuth consent abuse, or adversary-in-the-middle (AiTM) infrastructure. A tenant-aware lure does not need perfect technical sophistication if the social engineering is good enough to produce a valid login, a consent grant, or a trusted file open. Once that happens, session replay, mailbox rule manipulation, and internal impersonation become realistic next steps.

AI also helps generate content that looks like the real artefacts of work. That includes fake meeting notes, redlined contracts, onboarding forms, policy updates, executive briefs, and payment instructions that mirror the style of real business documents. The better the text looks, the more likely the victim is to treat the request as a normal part of collaboration.

Why document-based lures feel like normal work

Document-based lures are effective because they fit into everyday behaviour. Employees open shared files, review contract changes, approve expenses, comment on drafts, and respond to collaboration invites all day long. A request to read a policy update or check a redline does not feel unusual in the same way a random password reset email does.

Common lure types tend to follow ordinary business patterns:

• shared contracts
• payment requests
• policy updates
• HR forms
• meeting notes
• executive summaries
• security notices
• project briefs awaiting review

Each format has a built-in reason for urgency, and that makes it useful for attackers. A user who believes they are helping with a real process is less likely to pause and verify the source.

Collaboration platforms make this style of phishing even more persuasive. Shared documents, external invitations, comments, and access requests are all normal in Microsoft 365, Google Workspace, Slack, Teams, and shared drives. Attackers know that a file link can seem safer than a direct attachment, especially when it appears to come from a familiar tool or a known contact.

When document-based phishing succeeds, it usually does so because the victim believes they are doing legitimate work. That shift in mindset is powerful, because the user moves from evaluating danger to completing a task.

What the attack path looks like in practice

A typical attack starts with reconnaissance, where the attacker gathers whatever context they can from public sources and compromised data. They look for vendor names, reporting lines, document habits, meeting patterns, and the language people use to describe business processes. From there, AI helps shape the lure into something that sounds plausible in the target environment.

Example 1: financial services

Example 2: healthcare

Example 3: government or government-adjacent

The next step is the user action the attacker wants to trigger: opening a shared document, clicking a link, approving a request, entering credentials, granting OAuth consent, or downloading a file. Some campaigns are simple credential theft, while others aim for session hijacking, malware delivery, or business email compromise.

The chain is often shorter than people expect, but its impact can be wide. A single successful login can lead to mailbox access, document theft, internal impersonation, or lateral movement across cloud services. In organisations that rely on shared drives and external collaboration, one abused account can open a large amount of trusted content. That is why the document is rarely the end goal. It is the doorway.

Why security controls struggle to keep up

Traditional email filtering is still useful, but it was built to detect known-bad indicators such as malicious domains, suspicious attachments, and obvious spoofing. Tenant-specific lures often avoid those patterns by using legitimate cloud services, compromised accounts, or short-lived infrastructure that looks clean at first glance.

There is also a visibility problem in complex tenants. Security teams may not know what “normal” document-sharing behaviour looks like across every business unit, vendor relationship, and collaboration channel. Without good baselines, unusual access patterns can blend into ordinary activity.

That is why detection needs to move beyond content alone and into identity, collaboration, and behaviour. In modern environments, the signal often lives in:

• impossible travel or atypical geolocation
• unusual token use
• anomalous OAuth consent
• unexpected file-sharing spikes
• new external collaborators
• inbox rule creation
• access patterns that do not match the user’s role

For organisations managing regulated data, this is where NIST CSF 2.0, NIST SP 800-63B phishing-resistant authentication guidance, and ISO/IEC 27001 control discipline become useful in practice: not as branding, but as a way to build layered, evidence-based defence and response.

How organisations can defend cloud collaboration and identity

Defence needs to move beyond inbox filtering and into identity, collaboration, and behaviour. For many Australian organisations, that starts with tightening access controls, especially around MFA-resistant authentication, conditional access, and device posture checks.

A practical defensive stack should include:

1. Phishing-resistant authentication

1. OAuth and consent governance

1. External sharing controls

1. Behavioural analytics

1. Document-aware incident response

Where BCyber CyberSuite fits

• GRACE Portal brings the response into a Single Dashboard with Cross-product Analytics, Executive Reporting, Compliance Tracking, and Real-time Monitoring. That is especially useful when one lure touches email, identity, collaboration, and compliance teams at once.
• Paradigm supports GRC Automation, Synapse AI Engine, Risk Quantification, Policy Automation, and Compliance Mapping. In practice, that helps teams map phishing controls to APRA CPS 234, the Essential Eight, and ISO-aligned control objectives.
• Profiler adds Behavioral Analysis, Insider Threat Detection, Access Pattern Monitoring, Risk Scoring, and User Activity Analytics. That is exactly where tenant-specific lures tend to leave clues.
• CyberTriage is valuable when speed matters. Its 4-Minute Triage, Automated Response, Threat Containment, Forensic Analysis, and Incident Playbooks can shorten the time between suspicious click and containment.
• Perfect Storm is the right place to rehearse the scenario before it happens for real. Run a simulation of a shared-document lure against finance, HR, help desk, or executive assistants and test whether the organisation notices consent abuse, mailbox rules, or token replay.

BCyber also supports capability uplift through its 46-course library, which can be used to reinforce phishing awareness, cloud identity hygiene, and incident response readiness across business users and security teams.

For regulated organisations, this layered approach is not optional. It aligns with the practical expectations of APRA CPS 234, the ACSC Essential Eight, and the Privacy Act 1988, while also supporting stronger evidence for audits and board reporting.

Key takeaways

AI has made phishing more convincing, more adaptive, and more difficult to filter using old assumptions. Tenant-specific lures work because they imitate the target organisation’s own tools, language, and workflows rather than generic external brands. Document-based attacks succeed because they feel like ordinary work, which lowers suspicion at exactly the wrong moment.

The strongest defence is layered: phishing-resistant identity controls, tighter collaboration governance, behavioural detection, and training that reflects how modern attacks actually look. Organisations should assume attackers will keep borrowing trust from internal systems and should therefore monitor not just messages, but the actions that follow them.

For Australian businesses, the objective is simple: make a fake document, fake approval, or fake collaboration request expensive for the attacker and easy to verify for the user.

Verification note

This article includes BCyber platform and product information. The platform metrics cited above were verified against BCyber’s internal inventory as provided for this brief. External guidance should be checked against the latest regulator and standards body updates before implementation.

Further reading

• APRA CPS 234 Information Security: https://www.apra.gov.au/sites/default/files/2021-12/CPS%20234%20Information%20Security.pdf
• ACSC Essential Eight: https://www.cyber.gov.au/resources-business-and-government/essential-cyber-security/essential-eight
• ACSC phishing guidance: https://www.cyber.gov.au/topics/attacks-threats-and-crime/phishing
• NIST CSF 2.0: https://www.nist.gov/cyberframework
• NIST SP 800-63B Digital Identity Guidelines: https://pages.nist.gov/800-63-3/sp800-63b.html
• OAIC Notifiable Data Breaches scheme: https://www.oaic.gov.au/privacy/notifiable-data-breaches
• ISO/IEC 27001 overview: https://www.iso.org/standard/27001

BCyber recognition

• 2025 Finalist Cybersecurity Professional Of The Year — Karen Stephens, Co-Founder & CEO
• 2025 GRC Provider of the Year — Finalist
• 2025 AI Innovator - Cyber Security — Finalist