AI is industrializing social engineering and vulnerability discovery

At a glance

What this is: Expel’s Q1 2026 threat report argues that AI is being used to industrialize social engineering and speed up vulnerability discovery rather than to create entirely new attack classes.

Why it matters: For identity and security teams, the concern is that cheaper attacker workflows make trust, authentication, and access controls fail faster, especially where human verification and credential governance are already weak.

By the numbers:

• The group exfiltrated cryptocurrency wallet data from more than 2,700 developer systems in Q1.
• Their tracking infrastructure ingested wallets holding up to $12 million in assets over three months.

👉 Read Expel's Q1 2026 threat report on AI-driven social engineering and vulnerability discovery

Context

AI is changing attacker economics more than attacker physics. The core security problem is not that AI creates new laws of compromise, but that it lowers the cost of social engineering, malware development, and vulnerability discovery while making deceptive activity easier to scale across identity and access workflows.

This matters to identity security because many of the paths described in the report still depend on trust decisions, credential handling, and verification gaps. When fake companies, AI-generated personas, and AI-assisted code become routine attack infrastructure, identity verification, PAM, and human judgment all have to account for a more efficient adversary. That is now typical rather than exceptional.

For defenders, the practical question is whether controls are built to resist highly polished deception and faster exploit discovery, not just opportunistic phishing. The report’s examples fit a broader pattern seen across NHI abuse and social engineering: once the attacker can industrialize preparation, the control gap widens if verification and least privilege are still treated as static tasks.

Key questions

Q: How should security teams detect AI-generated social engineering that looks legitimate?

A: Security teams should combine behavioural analysis, workflow verification, and channel-based risk scoring. The goal is to detect whether a request fits historical communication patterns, not just whether it contains suspicious text. That approach is stronger than relying on links, attachments, or keyword triggers, which AI-generated fraud can easily avoid.

Q: Why does AI make identity and access controls easier to bypass?

A: AI lowers the cost of creating believable pretexts, fake personas, and polished communications, which lets attackers reach trust decisions more often. Once an organisation depends on informal approval, email confidence, or static onboarding checks, those controls become easier to manipulate. The risk is not new logic, but a much higher volume of convincing deception.

Q: What do security teams get wrong about AI exploit discovery?

A: Teams often assume exploit discovery remains a scarce human activity, but the article shows machine-speed discovery and chaining across real software surfaces. That changes how fast an exposed flaw can become a usable attack. The mistake is treating AI security as a future concern when the offensive capability is already operational.

Q: How should organisations respond when attackers can industrialize trust-building?

A: They should assume trust can be fabricated at scale and design verification accordingly. That means stronger onboarding controls, least-privilege defaults, tighter review of externally sourced identities, and faster internal discovery of software weaknesses. The goal is to make deception expensive again by forcing attackers to clear multiple independent checks.

Technical breakdown

How AI reduces the cost of industrialized social engineering

The report’s HexagonalRodent example shows AI being used as production infrastructure for deception. Fake company sites, generated headshots, fabricated leadership profiles, and polished job-offer lures can be created with far less labour than traditional tradecraft required. That does not make the attack technically novel, but it does make the operation more repeatable, more scalable, and easier to run like a campaign rather than a one-off scam. For defenders, the important shift is that trust indicators can now be manufactured at volume, which weakens simple visual vetting and template-based fraud checks.

Practical implication: strengthen identity verification and out-of-band validation for recruiting, vendor onboarding, and developer access workflows.

Why AI-assisted malware development changes defender assumptions

The report describes AI being used inside malware development, including code generation, prompt leakage, and self-review for detection resistance. The key mechanism is not autonomous malware, but faster iteration by attackers who can generate, test, and refine code with less specialist skill. That compresses the time between idea and deployable payload, which is especially dangerous when defenders rely on slow manual review cycles. The article also shows attackers trying to make their tooling less visible to analysis, which means code quality and signature-based detection are no longer reliable as standalone controls.

Practical implication: combine static analysis, behavioural detection, and sandboxing rather than relying on any single review step.

What AI-assisted vulnerability discovery means for exposure management

Anthropic’s Mythos research, as discussed in the report, points to a broader cost shift in vulnerability discovery. The vulnerabilities are not new, but the economics of finding and chaining them are changing. That matters because many exposure management models assume attacker effort stays relatively high and discovery remains bounded by human research capacity. If AI can surface weaknesses faster and cheaper, then obscurity stops being a meaningful cushion and closed-source code, internal code review, and attack-path analysis become more valuable as defensive discovery methods.

Practical implication: treat attacker discovery as a faster baseline and validate your own code and dependencies before someone else does.

Threat narrative

Attacker objective: The objective is to scale deceptive access, steal high-value digital assets, and reduce the effort required to find or weaponize weaknesses.

1. Entry begins with AI-generated social engineering that uses fake companies, job offers, and convincing personas to earn trust and initiate contact with targets.
2. Escalation follows when the attacker uses the relationship to deliver backdoored assessments or malicious tooling and then harvests wallet data, credentials, or other high-value assets.
3. Impact is achieved by turning AI-assisted preparation into scaled theft and faster exploit discovery, expanding the attacker’s reach while reducing the cost of each campaign.

NHI Mgmt Group analysis

AI has become attack infrastructure, not attack magic. The most important change in this report is economic, not technical. AI is making it cheaper to manufacture trust signals, produce convincing lures, and accelerate exploit research, which helps less skilled actors do more and skilled actors do it faster. For identity programmes, that means verification and access decisions now face industrialised deception rather than isolated fraud. The practitioner conclusion is that trust controls must assume scale.

Identity verification is now part of cyber resilience, not just fraud prevention. The fake company, fake persona, and backdoored assessment pattern shows how attacker tradecraft can run through recruitment, vendor engagement, and developer onboarding before any malware executes. This is where the boundary between identity governance and cyber defence becomes operational. NHI Mgmt Group’s position is that organisations should treat human verification controls as a security control surface, not an administrative step. The practitioner conclusion is that onboarding trust must be continuously validated.

Security through obscurity is losing its remaining value. The Mythos discussion reinforces a long-standing control gap: if discovery gets cheaper, hidden weaknesses become exposed faster. That does not mean every weakness is instantly exploitable, but it does mean defenders can no longer rely on attacker effort as a protective factor. The named concept here is discovery-cost compression, where lower research cost narrows the time between latent defect and active exploitation. The practitioner conclusion is that internal discovery and prioritisation must speed up.

The AI threat conversation is still being overstated in the wrong direction. The report is persuasive precisely because it rejects the idea of magical new malware while showing real operational change. That is the right frame for governance: model the acceleration of familiar abuse patterns, not fictional capability leaps. For IAM, PAM, and NHI governance, the implication is to focus on trust, privilege, and review cycles that assume a faster attacker. The practitioner conclusion is to align controls to attacker throughput, not headlines.

AI will widen the gap between what a control says and what it actually sees. When attackers can generate more believable identities, more code, and more reconnaissance at lower cost, shallow control measurement becomes less trustworthy. Mature programmes will need stronger evidence that identity checks, approvals, and detection logic still separate real from fabricated activity under pressure. The practitioner conclusion is to test whether controls fail gracefully when deception volume increases.

What this signals

The operational signal for practitioners is that identity controls now have to absorb higher volumes of manufactured trust, not just stolen credentials. As AI-assisted deception becomes cheaper, the weakest point is often the approval path, the onboarding workflow, or the unverified third party that appears legitimate enough to grant access.

Discovery-cost compression: attackers need less effort to find, test, and chain weaknesses, which narrows the window between latent exposure and active exploitation. That makes internal code review, asset visibility, and attack-path analysis more urgent than any claim that AI has invented new classes of vulnerability.

For identity programmes, the most useful response is to tie verification, privilege, and review to externally visible evidence and not to confidence in the requester. That aligns with the broader direction of zero trust and least privilege while acknowledging that AI now scales the attacker side of the equation.

For practitioners

• Harden identity verification for high-risk onboarding Require out-of-band validation for recruiters, contractors, and developers when offers, assessments, or vendor relationships create access pathways. Re-verify domain ownership, leadership identity, and communication channels before granting any privileged foothold.
• Review PAM assumptions around human trust Treat phishing-resistant authentication, step-up verification, and least privilege as controls against industrialized deception, not just stolen passwords. Prioritise privileged workflows that depend on email-based approval or informal identity checks.
• Add AI-era abuse cases to detection engineering Build detections for generated personas, duplicate hiring patterns, repeated assessment reuse, and rapid changes in social graph behaviour. Use them alongside malware and exfiltration alerts so the investigation starts before the payload lands.
• Shorten vulnerability discovery and review cycles Use internal code scanning, dependency review, and attack-path validation more aggressively because attacker discovery cost is falling. Prioritise externally reachable systems, public-facing credentials, and code paths that can be chained quickly.

Key takeaways

• AI is industrializing familiar attack patterns, especially social engineering and vulnerability discovery, rather than creating entirely new cyber physics.
• The report’s examples show that fake identities, generated personas, and AI-assisted code can compress attacker effort enough to outpace static verification models.
• Defenders should respond by strengthening identity proofing, reducing trust in informal approval paths, and accelerating internal discovery before attackers do.

Key terms

• AI-as-Infrastructure: A threat pattern where attackers use AI as part of the operational stack for deception, code generation, or research. The AI is not the payload itself. It is the production layer that lowers cost, increases scale, and makes familiar attack methods harder to spot.
• Discovery-Cost Compression: The reduction in time, effort, and expertise required to find exploitable weaknesses or produce convincing attack material. In practice, it shortens the attacker’s path from idea to action and weakens any defence that depends on obscurity, manual review, or slow analyst response.
• Industrialized Social Engineering: The use of repeatable, high-volume deception processes to manipulate people or access workflows. AI can help generate identities, content, and supporting infrastructure, making the campaign look credible at scale rather than relying on a few manually crafted lures.
• Identity proofing: The process of verifying that a person is who they claim to be before granting or restoring access. In higher-risk recovery paths, proofing can include stronger evidence checks such as government ID validation or liveness-based facial verification so the assurance level matches the sensitivity of the request.

What's in the full report

Expel's full Q1 Quarterly Threat Report covers the operational detail this post intentionally leaves for the source:

• The full HexagonalRodent investigation showing how AI-generated front companies and personas were operationalised across the campaign.
• The supporting threat-trend data from Expel Workbench™ that underpins the quarterly analysis.
• The Mythos discussion on how AI-assisted vulnerability discovery changes defender prioritisation and review timing.
• The two-part series context linking AI-as-bait with AI-as-infrastructure across Q1 2026.

👉 The full Expel report covers HexagonalRodent, Mythos, and the Q1 threat trends data behind the analysis.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management. It is designed for practitioners who need to connect identity controls to real operational risk across modern security programmes.