TL;DR: AI is lowering the barrier to industrialized social engineering and faster vulnerability discovery, according to Expel’s Q1 2026 threat report, with HexagonalRodent and Anthropic Mythos showing how attackers can scale fake personas, malware work, and exploit research. The real shift is not magical new capability, but cheaper, faster operations that stress identity controls, patching, and detection.
At a glance
What this is: Expel’s Q1 threat analysis says AI is being used to scale social engineering and accelerate vulnerability discovery rather than create entirely new attack classes.
Why it matters: IAM, PAM, and broader security teams should treat AI as attack infrastructure that amplifies existing weaknesses in trust, identity validation, and remediation speed.
By the numbers:
- The HexagonalRodent investigation exfiltrated cryptocurrency wallet data from more than 2,700 developer systems.
- Wallets holding up to $12 million in assets were ingested into the group’s tracking infrastructure over three months.
- Expel’s analysis identified six teams of roughly 31 operators.
Context
AI is not changing the mechanics of compromise so much as the economics of attack. In this article, the primary security problem is scale: AI makes social engineering cheaper to produce, easier to iterate, and more believable to run against human and technical trust controls.
That matters to identity security because the weakest point is often not authentication strength alone but the quality of the trust decision around the identity event. When fake companies, synthetic personas, and AI-assisted malware development can be assembled quickly, defenders have to assume higher-volume deception against human identity, developer access, and review workflows.
The second problem space is vulnerability discovery. AI-assisted research lowers the cost of finding flaws that previously remained hidden because exploitation required too much time, expertise, or manual chaining. That changes how organisations think about exposure windows and the value of their own code visibility.
Key questions
Q: What breaks when AI-powered social engineering is not in place?
A: The first thing that breaks is trust in informal verification. If employees can be persuaded by a convincing voice, face, or pretext, attackers can turn a single interaction into credential disclosure or privileged approval. The real failure is not the message itself. It is the absence of a stronger identity check before sensitive action is taken.
Q: Why do AI-assisted vulnerability tools change the exposure management model?
A: They shorten the time between discovery and validation, so teams can see exploitability while the environment is still changing. That makes prioritisation more accurate, but it also raises the need for strict scope controls, auditability, and evidence because machine-speed testing can expand faster than governance if left unchecked.
Q: How can security teams detect AI-mediated intrusion activity?
A: Look for rapid transitions between recon, credential use, and lateral movement, especially when the same identity is interacting with both AI tools and internal systems. Correlate prompts, tokens, and downstream actions. AI-mediated attacks often leave behavioural anomalies across systems rather than a single obvious malicious event.
A: They should do both, but identity hardening often limits the fastest spread while patching addresses the root entry point. If the exploit path already includes SSO, tokens, or privileged credentials, revocation and containment can reduce impact before the patch cycle completes.
Technical breakdown
AI-generated personas and social engineering scale
Modern social engineering depends on plausibility, repetition, and volume. AI lowers the cost of generating believable company sites, profiles, headshots, and written engagement at a speed that outpaces manual tradecraft. That does not make the deception technically novel, but it makes it industrial. For defenders, the important change is that trust signals can now be manufactured cheaply across multiple channels, including recruitment, developer outreach, and backdoored assessment workflows. Identity review, vendor validation, and human verification all become higher-friction when the attacker can sustain many more credible variants.
Practical implication: Treat trust decisions as attack surfaces and tighten verification for recruitment, partner onboarding, and developer-access requests.
AI-assisted malware development and signature evasion
The article’s malware discussion shows a different effect of AI: acceleration of coding, refinement, and counter-analysis. Attackers can use models to generate components, comment code, and then audit their own malware for signatures or analysis artefacts. That shifts the defensive problem from spotting obviously hand-written bad code to detecting operational patterns, prompt leakage, and generated content that still contains artefacts. AI does not remove operational mistakes, but it does reduce the skill barrier for producing usable tooling and iterating faster after detection. The result is more frequent, less expensive experimentation by actors who previously lacked the same development capacity.
Practical implication: Hunt for generated-code artefacts and use layered detection, not signature-only controls, against AI-assisted malware.
AI-assisted vulnerability discovery changes exposure economics
Vulnerability discovery has always been constrained by time, expertise, and the cost of chaining weaknesses into a usable exploit path. AI compresses that cost curve by making search, triage, and hypothesis testing cheaper. The article’s key point is not that vulnerabilities suddenly appear, but that hidden flaws become easier to surface and prioritise. That matters most where organisations rely on obscurity, slow manual review, or immature secure development practices. Closed-source defenders may still have an advantage if they use AI against their own code before adversaries reverse-engineer it from binaries or observable behaviour.
Practical implication: Assume discovery speed will increase and shorten the interval between code introduction, vulnerability finding, and remediation.
Threat narrative
Attacker objective: The objective is to scale developer compromise, steal valuable wallet data, and convert AI-enabled deception into repeatable operational theft.
- Entry begins with AI-assisted social engineering that uses fake companies, AI-generated personas, and targeted job offers to lure developers into contact with the attacker.
- Credential or data access follows when victims interact with backdoored skills assessments and the attackers exfiltrate wallet data and related developer information from compromised systems.
- Impact is industrialised theft and broader operational scaling, with attackers using AI to maintain the deception, support malware development, and improve their own tooling and targeting.
Breaches seen in the wild
- Bybit hack 2025: Hijacked AWS session tokens from a Safe{Wallet} developer's laptop let attackers alter signing code and steal about $1.5 billion from Bybit.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
AI has become attack infrastructure, not just an attacker tool. The strongest reading of this article is that AI now lowers production costs across the whole attack lifecycle, from persona creation to malware iteration. That matters because scale changes risk more than novelty does. Security teams should evaluate whether their controls are built to resist high-volume, low-cost deception rather than isolated advanced adversaries.
Identity governance now has to absorb synthetic trust pressure. Fake companies, fake recruiters, and AI-generated headshots target the human decision layer that sits around IAM, not just the authentication event itself. The practical issue is trust qualification before access is ever granted. Organisations that only harden login will still be exposed if their intake, validation, and exception handling remain easy to spoof.
AI-assisted vulnerability discovery compresses the useful life of obscurity. Mythos-style tooling does not invent new weaknesses, but it makes the search for existing ones cheaper and faster. That is a governance problem as much as a technical one because exposure windows shrink when discovery economics change. The practitioner response is to treat internal code review, validation, and patch acceleration as part of the same risk model.
Developer trust is now a security boundary, not a soft control. The HexagonalRodent pattern shows that targeted work on developers can blend social engineering, access harvesting, and platform abuse into one campaign. In NHI-adjacent terms, that means service access, assessments, and ephemeral interactions can all be used as entry points if the surrounding trust model is weak. Teams should treat developer workflows as privileged pathways, not peripheral business processes.
What this signals
AI industrialisation is most dangerous when it meets weak trust choreography. The practical lesson for identity and security teams is that higher-quality deception can outpace controls that only check credentials at the point of login. Recruiting funnels, contractor onboarding, developer assessments, and partner access all need stronger proof before a trust decision becomes an access decision.
Vulnerability management now competes with machine-speed discovery pressure. AI makes it more plausible that weaknesses will be found sooner and at lower cost, which means traditional remediation queues become a governance liability. Security programmes should narrow the gap between internal code review, testing, and patch execution before attackers use the same economics against them.
For practitioners
- Harden developer recruitment and assessment workflows Require out-of-band verification for recruiting contacts, assessment links, and any request that moves a candidate into code execution or file exchange.
- Add validation for AI-generated persona risk Screen for synthetic company sites, copied bios, inconsistent employment history, and unusually polished contact profiles before allowing internal engagement.
- Shorten vulnerability discovery and remediation cycles Use AI-assisted review internally to search your own codebase first, then map findings into a faster triage and patch workflow.
- Strengthen identity checks on high-risk developer access Treat developer onboarding, repository access, and assessment environments as privileged entry points that need stronger authentication and approval controls.
Key takeaways
- AI is reducing the cost of believable social engineering and turning trust validation into a higher-volume security problem.
- The HexagonalRodent case shows how synthetic personas, fake companies, and backdoored assessments can be combined into industrialised attack operations.
- Defenders need to treat developer workflows, internal code review, and remediation speed as linked controls because AI compresses both deception and vulnerability discovery timelines.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK addresses the attack and risk surface, while NIST CSF 2.0, CIS Controls v8 and NIST SP 800-53 Rev 5 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| MITRE ATT&CK | TA0006; TA0007; TA0009 — Credential Access; Discovery; Collection | The article centers on AI-assisted social engineering, discovery, and data theft patterns. |
| Recommendation — Map AI-assisted tradecraft to ATT&CK tactics and tune detections for credential access, discovery, and collection. | ||
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | Identity validation and access decisions are the defensive choke point in the article. |
| Recommendation — Apply PR.AA-05 to tighten approval and entitlement checks before developer access is granted. | ||
| CIS Controls v8 | CIS-5 — Account Management | The article highlights the need to govern who can gain and retain access in high-trust workflows. |
| Recommendation — Use CIS-5 to review account creation, approval, and removal paths for sensitive developer and partner access. | ||
| NIST SP 800-53 Rev 5 | IA-5 — Authenticator Management | AI-driven deception raises the value of stronger authenticator lifecycle control for access entry points. |
| AC-2 — Account Management | Account onboarding and ongoing account control are directly implicated by social engineering and access abuse. | |
| Recommendation — Enforce IA-5 to manage authenticator issuance, rotation, and revocation across privileged workflows. Use AC-2 to tighten account provisioning, review, and deprovisioning for developer-facing systems. | ||
Key terms
- AI-powered social engineering: AI-powered social engineering is the use of generated text, voice, video, or interface content to manipulate a target into taking an unsafe action. The goal is not just deception, but trust transfer, where the attacker convinces a legitimate identity holder to approve, disclose, or execute something harmful.
- Synthetic Persona: A synthetic persona is a fabricated identity designed to look credible to a real person or system. In fraud settings, it combines profile data, images, dialogue, and behaviour to build trust. The risk is not just false registration, but the ability to sustain deception long enough to trigger real-world harm.
- Vulnerability discovery economics: The cost, time, and expertise required to find exploitable software weaknesses. AI changes these economics by making search and triage cheaper, which shortens the safe period in which flaws remain undiscovered by adversaries.
- Trust validation workflow: The sequence of checks used to decide whether a person, company, or request should be believed before access is granted. It often sits outside formal IAM controls, yet it strongly influences whether identity decisions are made safely.
Deepen your knowledge
NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, workload identity security, and secrets management. It is suited to practitioners who need to connect identity controls to the wider threat patterns described in this analysis.
Published by the NHIMG editorial team on August 18, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org