Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
Research demo

Lovable's vibe-coding platform was prompted into generating, hosting and serving live credential-harvesting phishing pages on its own subdomain

AIR-2025-0022 AIR record · 2025-04-lovable-vibescamming-phishing-hosting
From the supporting source
As a purpose-built tool for creating and deploying web apps, its capabilities line up perfectly with every scammer's wishlist,labs.guard.io
The record
Disclosed
2025-04-09
Occurred
2025-03
Vendor
Lovable
Product
Lovable
Risk domain
Governance › Policy drift
System type
Generative model
Who acted
Agent acted — the platform generated, hosted and served live phishing pages
Attack surface
Agent framework
Autonomy level
Semi-autonomous
Initial vector
Jailbreak guardrail bypass
Impact type
Ecosystem abuse
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI01 – Agent Goal Hijack
MITRE ATLAS
AML.T0010 – AI Supply Chain Compromise
NIST AI RMF
GOVERN 1.5
EU AI Act
—
ISO/IEC
TR 24028; 23894
AIUC-1
E. Accountability
Scope
agent failure
Mitigation
Implement version pinning for models and prompts; Monitor for behavioral drift.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
any Lovable user; generated pages were live on lovable.app subdomains with real-time exfiltration to Telegram and Firebase
Capability abused
code_generationapp_deploymenthostingdata_exfiltration_integration
Defense & remediation
Recorded defense
platform safety guardrails
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
Guardio disclosed to vendors; Lovable scored lowest on the VibeScamming benchmark
Taxonomy & references
OWASP ASI01 Agent Goal Hijackvibescammingjailbreakbuild_and_host_in_onereputable_domain_abuselowest_benchmark_score
Curator notes
Guardio Labs' VibeScamming Benchmark v1.0. class is research_demo — researchers built the campaigns, no victims confirmed — but the agent did the whole kill chain end to end: wrote the page, deployed it live on a domain that inherits the platform's reputation, and wired up exfiltration. Build-and-host agents collapse the gap between generating an attack and operating one. AI Incident Database incident 1016.
Cite as
Enkrypt AI. "Lovable's vibe-coding platform was prompted into generating, hosting and serving live credential-harvesting phishing pages on its own subdomain". AIR, AIR-2025-0022. https://agentincident.vercel.app/i/AIR-2025-0022/ (accessed 2026-10-19).