5,000 vibe-coded apps just proved shadow AI is the new S3 bucket crisis
Our take
The security industry has spent years building sophisticated frameworks to protect servers, cloud infrastructure, and corporate endpoints. Yet as the RedAccess research reveals, a new category of enterprise assets has emerged outside those protections entirely: applications built by employees over a weekend using vibe coding platforms, deployed on public URLs, and connected to live databases containing sensitive corporate data. The scale is striking—380,000 publicly accessible assets across Lovable, Base44, Replit, and Netlify, with roughly 5,000 containing sensitive information spanning healthcare records, financial data, and customer PII. This is not a theoretical risk. It is an active, quantified exposure that security teams have no visibility into, and the cost implications are already materializing. IBM's 2025 Cost of a Data Breach Report found that shadow AI breaches added an average of $670,000 to incident costs, pushing the shadow AI breach average to $4.63 million. The pattern extends beyond vibe-coded apps themselves. Recent research on Claude Code, Copilot and Codex all got hacked. Every attacker went for the credential, not the model demonstrates that AI development tools are being targeted across multiple vectors, while the Vercel breach exposes the OAuth gap most security teams cannot detect, scope or contain illustrates how quickly a single unauthorized tool adoption can cascade into a supply chain compromise. Together, these findings paint a picture of security architectures that were designed for a different era of enterprise computing.
The core problem is structural, not procedural. Traditional asset discovery tools were built to find servers, containers, and cloud instances—finite, trackable infrastructure with identifiable fingerprints. They have no mechanism for discovering a marketing configurator that a product manager built on Lovable, connected to a Supabase database holding live customer records, and shared with external contractors through a public URL that Google indexed within hours. The platforms themselves compound the issue: privacy defaults make apps publicly accessible unless users manually opt for privacy, and the assumption embedded in that design choice is that the platform handles security. It does not. The CVE-2025-48757 documentation found insufficient or missing Row-Level Security policies in Lovable-generated Supabase projects, meaning the AI generated the database layer but not the access controls that should have restricted who could read the data. This is the critical insight: AI generates syntactically correct code that lacks awareness of broader system architecture and nuanced business rules, a defect category Gartner forecasts will increase software defects by 2,500% by 2028.
What makes this particularly challenging is that the exposure happens at AI-generation speed while security review remains a human process. Escape.tech's October 2025 scan found over 2,000 high-impact vulnerabilities and 175 instances of personal data exposure across 5,600 vibe-coded applications, with every vulnerability residing in a live production system discoverable within hours. The velocity mismatch is stark: employees can deploy functioning applications in minutes, but security teams have no inventory of what was built, what data it connects to, or whether authentication is required. The detection challenge runs deeper than most organizations realize. These apps deploy on platform subdomains that rotate frequently and sit behind CDN layers that mask origin infrastructure. Conventional SIEM and endpoint telemetry generate limited signals for applications that exist in the gap between network visibility and application inventory—a gap most security stacks were never architected to cover.
The question for security leaders is not whether vibe-coded apps exist inside their perimeter. The question is how many, holding what data, visible to whom. The RedAccess findings suggest the answer, for most organizations, is worse than anyone in the C-suite currently knows. Organizations that treat this as a policy problem will write memos that gather dust. Those that treat it as an architecture problem will deploy discovery scanning across the major vibe coding domains, require pre-deployment security review, extend AppSec pipelines to citizen-built applications, and add these platforms to DLP rules. The structural failure is the same across the ecosystem: security review happens after deployment or not at all. Identity systems track human users and service accounts, but they do not track the Lovable app a sales operations analyst deployed last Tuesday connected to a live CRM database. The organizations that start scanning this week will find them. The ones that wait will read about themselves next.
Most enterprise security programs were built to protect servers, endpoints, and cloud accounts. None of them was built to find a customer intake form that a product manager vibe coded on Lovable over a weekend, connected to a live Supabase database, and deployed on a public URL indexed by Google. That gap now has a price tag.
New research from Israeli cybersecurity firm RedAccess quantifies the scale. The firm discovered 380,000 publicly accessible assets, including applications, databases, and related infrastructure, built with vibe coding tools from Lovable, Base44, and Replit, as well as deployment platform Netlify. Roughly 5,000 of those assets, about 1.3%, contained sensitive corporate information. CEO Dor Zvi said his team found the exposure while researching shadow AI for customers. Axios independently verified multiple exposed apps, and Wired confirmed the findings separately.
Among the verified exposures: a shipping company app detailed which vessels were expected at which ports. An internal health company application listed active clinical trials across the U.K. Full, unredacted customer service conversations for a British cabinet supplier sat on the open web. Internal financial information for a Brazilian bank was accessible to anyone who found the URL.
The exposed data also included patient conversations at a children’s long-term care facility, hospital doctor-patient summaries, incident response records at a security company, and ad purchasing strategies. Depending on jurisdiction and the data involved, the healthcare and financial exposures may trigger regulatory obligations under HIPAA, UK GDPR, or Brazil’s LGPD.
RedAccess found phishing sites built on Lovable that impersonated Bank of America, FedEx, Trader Joe’s, and McDonald’s. Lovable said it had begun investigating and removing the phishing sites.
The defaults are the problem
Privacy settings on several vibe coding platforms make apps publicly accessible unless users manually switch them to private. Many of these applications get indexed by Google and other search engines. Anyone can stumble across them. Zvi put it plainly: “I don’t think it’s feasible to educate the whole world around security. My mother is [vibe coding] with Lovable, and no offense, but I don’t think she will think about role-based access.”
This is not an isolated finding
In October 2025, Escape.tech scanned 5,600 publicly available vibe-coded applications and found more than 2,000 high-impact vulnerabilities, over 400 exposed secrets including API keys and access tokens, and 175 instances of personal data exposure containing medical records and bank account numbers. Every vulnerability Escape found was in a live production system, discoverable within hours. The full report documents the methodology. Escape separately raised an $18 million Series A led by Balderton in March 2026, citing the security gap opened by AI-generated code as a core market thesis.
Gartner’s “Predicts 2026” report forecasts that by 2028, prompt-to-app approaches adopted by citizen developers will increase software defects by 2,500%. Gartner identifies a new class of defect where AI generates code that is syntactically correct but lacks awareness of broader system architecture and nuanced business rules. The remediation costs for these deep contextual bugs will consume budgets previously allocated to innovation.
Shadow AI is the multiplier
IBM’s 2025 Cost of a Data Breach Report found that 20% of organizations experienced breaches linked to shadow AI. Those incidents added $670,000 to the average breach cost, pushing the shadow AI breach average to $4.63 million. Among organizations that reported AI-related breaches, 97% lacked proper access controls. And 63% of breached organizations had no AI governance policy in place.
Shadow AI breaches disproportionately exposed customer personally identifiable information at 65%, compared to 53% across all breaches, and affected data distributed across multiple environments 62% of the time. Only 34% of organizations with AI governance policies performed regular audits for unsanctioned AI tools. VentureBeat’s shadow AI research estimated that actively used shadow apps could more than double by mid-2026. Cyberhaven data found 73.8% of ChatGPT workplace accounts in enterprise environments were unauthorized.
What to do first
The audit framework below gives CISOs a starting point for triaging vibe-coded app risk across five domains.
Domain | Current State (Most Orgs) | Target State | First Action |
Discovery | No visibility into vibe-coded apps | Automated scanning of vibe coding platform domains | Run DNS + certificate transparency scan for Lovable, Replit, Base44, and Netlify subdomains tied to corporate assets |
Authentication | Platform defaults (public by default) | SSO/SAML integration required before deployment | Block unauthenticated apps from accessing internal data sources |
Code scanning | Zero coverage for citizen-built apps | Mandatory SAST/DAST before production | Extend the existing AppSec pipeline to cover vibe-coded deployments |
Data loss prevention | No DLP coverage for vibe coding domains | DLP policies covering Lovable, Replit, Base44, Netlify | Add vibe coding platform domains to existing DLP rules |
Governance | No AI usage policy or shadow AI detection | AI governance policy with regular audits for unsanctioned tools | Publish an acceptable-use policy for AI coding tools with a pre-deployment review gate |
The CISO who treats this as a policy problem will write a memo. The CISO who treats this as an architecture problem will deploy discovery scanning across the four largest vibe coding domains, require pre-deployment security review, extend the existing AppSec pipeline to citizen-built apps, and add those domains to DLP rules before the next board meeting. One of those CISOs avoids the next headline.
The vibe coding exposure RedAccess documented is not a separate problem from shadow AI. It is shadow AI's production layer. Employees build internal tools on platforms that default to public, skip authentication, and never appear on any asset inventory, which means the applications stay invisible to security teams until a breach surfaces or a reporter finds them first. Traditional asset discovery tools were designed to find servers, containers, and cloud instances. They have no way to find a marketing configurator that a product manager built on Lovable over a weekend, connected to a Supabase database holding live customer records, and shared with three external contractors through a public URL that Google indexed within hours.
The detection challenge runs deeper than most security teams realize. Vibe-coded apps deploy on platform subdomains that rotate frequently and often sit behind CDN layers that mask origin infrastructure. Organizations running mature, secure web gateways, CASB, or DNS logging can detect employee access to these domains. But detecting access is not the same as inventorying what was deployed, what data it holds, or whether it requires authentication. Without explicit monitoring of the major vibe coding platforms, the apps themselves generate a limited signal in conventional SIEM or endpoint telemetry. They exist in a gap between network visibility and application inventory that most security stacks were never architected to cover.
The platform responses tell the story
Replit CEO Amjad Masad said RedAccess gave his company only 24 hours before going to the press. Base44 (via Wix) and Lovable both said RedAccess did not include the URLs or technical specifics needed to verify the findings. None of the platforms denied that the exposed applications existed.
Wiz Research separately discovered in July 2025 that Base44 contained a platform-wide authentication bypass. Exposed API endpoints allowed anyone to create a verified account on private apps using nothing more than a publicly visible app_id. The flaw meant that showing up to a locked building and shouting a room number was enough to get the doors open. Wix fixed the vulnerability within 24 hours after Wiz reported it, but the incident exposed how thin the authentication layer is on platforms where millions of apps are being built by users who assume the platform handles security for them.
The pattern is consistent across the vibe coding ecosystem. CVE-2025-48757 documented insufficient or missing Row-Level Security policies in Lovable-generated Supabase projects. Certain queries skipped access checks entirely, exposing data across more than 170 production applications. The AI generated the database layer. It did not generate the security policies that should have restricted who could read the data. Lovable disputes the CVE classification, stating that individual customers accept responsibility for protecting their application data. That dispute itself illustrates the core tension: platforms that market to nontechnical builders are shifting security responsibility to users who do not know it exists.
What this means for security teams
The RedAccess findings complete the picture. Professional agents face credential theft on one layer. Citizen platforms face data exposure on the other. The structural failure is the same. Security review happens after deployment or not at all. Identity and access management systems track human users and service accounts. They do not track the Lovable app a sales operations analyst deployed last Tuesday, connected to a live CRM database, and shared with three external contractors via a public URL.
Nobody asks whether the database policies restrict who can read the data or whether the API endpoints require authentication. When those questions go unasked at AI-generation speed, the exposure scales faster than any human review process can match. The question for security leaders is not whether vibe-coded apps are inside their perimeter. The question is how many, holding what data, visible to whom. The RedAccess findings suggest the answer, for most organizations, is worse than anyone in the C-suite currently knows. The organizations that start scanning this week will find them. The ones that wait will read about themselves next.
Read on the original site
Open the publisher's page for the full experience
Related Articles
- Claude Code, Copilot and Codex all got hacked. Every attacker went for the credential, not the model.On March 30, BeyondTrust proved that a crafted GitHub branch name could steal Codex’s OAuth token in cleartext. OpenAI classified it Critical P1. Two days later, Anthropic’s Claude Code source code spilled onto the public npm registry, and within hours, Adversa found Claude Code silently ignored its own deny rules once a command exceeded 50 subcommands. These were not isolated bugs. They were the latest in a nine-month run: six research teams disclosed exploits against Codex, Claude Code, Copilot, and Vertex AI, and every exploit followed the same pattern. An AI coding agent held a credential, executed an action, and authenticated to a production system without a human session anchoring the request. The attack surface was first demonstrated at Black Hat USA 2025, when Zenity CTO Michael Bargury hijacked ChatGPT, Microsoft Copilot Studio, Google Gemini, Salesforce Einstein and Cursor with Jira MCP on stage with zero clicks. Nine months later, those credentials are what attackers reached. Merritt Baer, CSO at Enkrypt AI and former Deputy CISO at AWS, named the failure in an exclusive VentureBeat interview. “Enterprises believe they’ve ‘approved’ AI vendors, but what they’ve actually approved is an interface, not the underlying system.” The credentials underneath the interface are the breach. Codex, where a branch name stole GitHub tokens BeyondTrust researcher Tyler Jespersen, with Fletcher Davis and Simon Stewart, found Codex cloned repositories using a GitHub OAuth token embedded in the git remote URL. During cloning, the branch name parameter flowed unsanitized into the setup script. A semicolon and a backtick subshell turned the branch name into an exfiltration payload. Stewart added the stealth. By appending 94 Ideographic Space characters (Unicode U+3000) after “main,” the malicious branch looked identical to the standard main branch in the Codex web portal. A developer sees “main.” The shell sees curl exfiltrating their token. OpenAI classified it Critical P1 and shipped full remediation by February 5, 2026. Claude Code, where two CVEs and a 50-subcommand bypass broke the sandbox CVE-2026-25723 hit Claude Code’s file-write restrictions. Piped sed and echo commands escaped the project sandbox because command chaining was not validated. Patched in 2.0.55. CVE-2026-33068 was subtler. Claude Code resolved permission modes from .claude/settings.json before showing the workspace trust dialog. A malicious repo set permissions.defaultMode to bypassPermissions. The trust prompt never appeared. Patched in 2.1.53. The 50-subcommand bypass landed last. Adversa found that Claude Code silently dropped deny-rule enforcement once a command exceeded 50 subcommands. Anthropic’s engineers had traded security for speed and stopped checking after the fiftieth. Patched in 2.1.90. “A significant vulnerability in enterprise AI is broken access control, where the flat authorization plane of an LLM fails to respect user permissions,” wrote Carter Rees, VP of AI and Machine Learning at Reputation and a member of the Utah AI Commission. The repository decided what permissions the agent had. The token budget decided which deny rules survived. Copilot, where a pull request description and a GitHub issue both became root Johann Rehberger demonstrated CVE-2025-53773 against GitHub Copilot with Markus Vervier of Persistent Security as co-discoverer. Hidden instructions in PR descriptions triggered Copilot to flip auto-approve mode in .vscode/settings.json. That disabled all confirmations and granted unrestricted shell execution across Windows, macOS, and Linux. Microsoft patched it in the August 2025 Patch Tuesday release. Then, Orca Security cracked Copilot inside GitHub Codespaces. Hidden instructions in a GitHub issue manipulated Copilot into checking out a malicious PR with a symbolic link to /workspaces/.codespaces/shared/user-secrets-envs.json. A crafted JSON $schema URL exfiltrated the privileged GITHUB_TOKEN. Full repository takeover. Zero user interaction beyond opening the issue. Mike Riemer, CTO at Ivanti, framed the speed dimension in a VentureBeat interview: “Threat actors are reverse engineering patches within 72 hours. If a customer doesn’t patch within 72 hours of release, they’re open to exploit.” Agents compress that window to seconds. Vertex AI, where default scopes reached Gmail, Drive and Google’s own supply chain Unit 42 researcher Ofir Shaty found that the default Google service identity attached to every Vertex AI agent had excessive permissions. Stolen P4SA credentials granted unrestricted read access to every Cloud Storage bucket in the project and reached restricted, Google-owned Artifact Registry repositories at the core of the Vertex AI Reasoning Engine. Shaty described the compromised P4SA as functioning like a "double agent," with access to both user data and Google's own infrastructure. VentureBeat defense grid Security requirement Defense shipped Exploit path The gap Sandbox AI agent execution Codex runs tasks in cloud containers; token scrubbed during agent runtime. Token present during cloning. Branch-name command injection executed before cleanup. No input sanitization on container setup parameters. Restrict file system access Claude Code sandboxes writes via accept-edits mode. Piped sed/echo escaped sandbox (CVE-2026-25723). Settings.json bypassed trust dialog (CVE-2026-33068). 50-subcommand chain dropped deny-rule enforcement. Command chaining not validated. Settings loaded before trust. Deny rules truncated for performance. Block prompt injection in code context Copilot filters PR descriptions for known injection patterns. Hidden injections in PRs, README files, and GitHub issues triggered RCE (CVE-2025-53773 + Orca RoguePilot). Static pattern matching loses to embedded prompts in legitimate review and Codespaces flows. Scope agent credentials to least privilege Vertex AI Agent Engine uses P4SA service agent with OAuth scopes. Default scopes reached Gmail, Calendar, Drive. P4SA credentials read every Cloud Storage bucket and Google’s Artifact Registry. OAuth scopes non-editable by default. Least privilege violated by design. Inventory and govern agent identities No major AI coding agent vendor ships agent identity discovery or lifecycle management. Not attempted. Enterprises do not inventory AI coding agents, their credentials, or their permission scopes. AI coding agents are invisible to IAM, CMDB, and asset inventory. Zero governance exists. Detect credential exfiltration from agent runtime Codex obscures tokens in web portal view. Claude Code logs subcommands. Tokens visible in cleartext inside containers. Unicode obfuscation hid exfil payloads. Subcommand chaining hid intent. No runtime monitoring of agent network calls. Log truncation hid the bypass. Audit AI-generated code for security flaws Anthropic launched Claude Code Security (Feb 2026). OpenAI launched Codex Security (March 2026). Both scan generated code. Neither scans the agent’s own execution environment or credential handling. Code-output security is not agent-runtime security. The agent itself is the attack surface. Every exploit targeted runtime credentials, not model output Every vendor shipped a defense. Every defense was bypassed. The Sonar 2026 State of Code Developer Survey found 25% of developers use AI agents regularly, and 64% have started using them. Veracode tested more than 100 LLMs and found 45% of generated code samples introduced OWASP Top 10 flaws, a separate failure that compounds the runtime credential gap. CrowdStrike CTO Elia Zaitsev framed the rule in an exclusive VentureBeat interview at RSAC 2026: collapse agent identities back to the human, because an agent acting on your behalf should never have more privileges than you do. Codex held a GitHub OAuth token scoped to every repository the developer authorized. Vertex AI’s P4SA read every Cloud Storage bucket in the project. Claude Code traded deny-rule enforcement for token budget. Kayne McGladrey, an IEEE Senior Member who advises enterprises on identity risk, made the same diagnosis in an exclusive interview with VentureBeat. "It uses far more permissions than it should have, more than a human would, because of the speed of scale and intent." Riemer drew the operational line in an exclusive VentureBeat interview. "It becomes, I don't know you until I validate you." The branch name talked to the shell before validation. The GitHub issue talked to Copilot before anyone read it. Security director action plan Inventory every AI coding agent (CIEM). Codex, Claude Code, Copilot, Cursor, Gemini Code Assist, Windsurf. List the credentials and OAuth scopes each received at setup. If your CMDB has no category for AI agent identities, create one. Audit OAuth scopes and patch levels. Upgrade Claude Code to 2.1.90 or later. Verify Copilot's August 2025 patch. Migrate Vertex AI to the bring-your-own-service-account model. Treat branch names, pull request descriptions, GitHub issues, and repo configuration as untrusted input. Monitor for Unicode obfuscation (U+3000), command chaining over 50 subcommands, and changes to .vscode/settings.json or .claude/settings.json that flip permission modes. Govern agent identities the way you govern human privileged identities (PAM/IGA). Credential rotation. Least-privilege scoping. Separation of duties between the agent that writes code and the agent that deploys it. CyberArk, Delinea, and any PAM platform that accepts non-human identities can onboard agent OAuth credentials today; Gravitee's 2026 survey found only 21.9% of teams have done it. Validate before you communicate. "As long as we trust and we check and we validate, I'm fine with letting AI maintain it," Riemer said. Before any AI coding agent authenticates to GitHub, Gmail, or an internal repository, verify the agent's identity, scope, and the human session it is bound to. Ask each vendor in writing before your next renewal. "Show me the identity lifecycle management controls for the AI agent running in my environment, including credential scope, rotation policy, and permission audit trail." If the vendor cannot answer, that is the audit finding. The governance gap in three sentences Most CISOs inventory every human identity and have zero inventory of the AI agents running with equivalent credentials. No IAM framework governs human privilege escalation and agent privilege escalation with the same rigor. Most scanners track every CVE but cannot alert when a branch name exfiltrates a GitHub token through a container that developers trust by default. Zaitsev's advice to RSAC 2026 attendees was blunt: you already know what to do. Agents just made the cost of not doing it catastrophic.
- Vercel breach exposes the OAuth gap most security teams cannot detect, scope or containOne employee at Vercel adopted an AI tool. One employee at that AI vendor got hit with an infostealer. That combination created a walk-in path to Vercel’s production environments through an OAuth grant that nobody had reviewed. Vercel, the cloud platform behind Next.js and its millions of weekly npm downloads, confirmed on Sunday that attackers gained unauthorized access to internal systems. Mandiant was brought in. Law enforcement was notified. Investigations remain active. An update on Monday confirmed that Vercel collaborated with GitHub, Microsoft, npm, and Socket to verify that no Vercel npm packages were compromised. Vercel also announced it is now defaulting environment variable creation to “sensitive.” Next.js, Turbopack, AI SDK, and all Vercel-published npm packages remain uncompromised after a coordinated audit with GitHub, Microsoft, npm, and Socket. Context.ai was the entry point. OX Security’s analysis found that a Vercel employee installed the Context.ai browser extension and signed into it using a corporate Google Workspace account, granting broad OAuth permissions. When Context.ai was breached, the attacker inherited that employee’s Workspace access, pivoted into Vercel environments, and escalated privileges by sifting through environment variables not marked as “sensitive.” Vercel’s bulletin states that variables marked sensitive are stored in a manner that prevents them from being read. Variables without that designation were accessible in plaintext through the dashboard and API, and the attacker used them as the escalation path. CEO Guillermo Rauch described the attacker as “highly sophisticated and, I strongly suspect, significantly accelerated by AI.” Jaime Blasco, CTO of Nudge Security, independently surfaced a second OAuth grant tied to Context.ai’s Chrome extension, matching the client ID from Vercel’s published IOC to Context.ai’s Google account before Rauch’s public statement. The Hacker News reported that Google removed Context.ai’s Chrome extension from the Chrome Web Store on March 27. Per The Hacker News and Nudge Security, that extension embedded a second OAuth grant enabling read access to users’ Google Drive files. Patient zero. A Roblox cheat and a Lumma Stealer infection Hudson Rock published forensic evidence on Monday, reporting that the breach origin traces to a February 2026 Lumma Stealer infection on a Context.ai employee’s machine. According to Hudson Rock, browser history showed the employee downloading Roblox auto-farm scripts and game exploit executors. Harvested credentials included Google Workspace logins, Supabase keys, Datadog tokens, Authkit credentials, and the support@context.ai account. Hudson Rock identified the infected user as a core member of “context-inc,” Context.ai’s tenant on the Vercel platform, with administrative access to production environment variable dashboards. Context.ai published its own bulletin on Sunday (updated Monday), disclosing that the breach affects its deprecated AI Office Suite consumer product, not its enterprise Bedrock offering (Context.ai’s agent infrastructure product, unrelated to AWS Bedrock). Context.ai says it detected unauthorized access to its AWS environment in March, hired CrowdStrike to investigate, and shut down the environment. Its updated bulletin then disclosed that the scope was broader than initially understood: the attacker also compromised OAuth tokens for consumer users, and one of those tokens opened the door to Vercel’s Google Workspace. Dwell time is the detail that should concern security directors. Nearly a month separated Context.ai’s March detection from the Vercel disclosure on Sunday. A separate Trend Micro analysis references an intrusion beginning as early as June 2024 — a finding that, if confirmed, would extend the dwell time to roughly 22 months. VentureBeat could not independently reconcile that timeline with Hudson Rock's February 2026 dating; Trend Micro did not respond to a request for comment before publication. Where detection goes blind Security directors can use this table to benchmark their own detection stack against the four-hop kill chain this breach exploited. Kill Chain Hop What Happened Who Should Detect Typical Coverage Gap 1. Infostealer on employee device Context.ai employee downloaded Roblox cheat scripts; Lumma Stealer harvested Workspace creds, Supabase/Datadog/Authkit keys. EDR on endpoint; credential exposure monitoring. Low. Device likely under-monitored. No stealer log monitoring at most orgs. Most enterprises do not subscribe to infostealer intelligence feeds or correlate stealer logs against employee email domains. 2. AWS compromise at Context.ai Attacker used harvested credentials to access Context.ai’s AWS. Detected in March. Context.ai cloud security; AWS CloudTrail. Partially detected. Context.ai stopped AWS access but missed OAuth token exfiltration. Initial investigation did not identify OAuth token exfiltration. Scope was underestimated until Vercel disclosure. 3. OAuth token theft into Vercel Workspace Compromised OAuth token used to access a Vercel employee’s Google Workspace. Employee had granted “Allow All” permissions via Chrome extension. Google Workspace audit logs; OAuth app monitoring; CASB. Very low. Most orgs do not monitor third-party OAuth token usage patterns. No approval workflow intercepted the grant. No anomaly detection on OAuth token use from a compromised third party. This is the hop no one saw. 4. Lateral movement into Vercel production Attacker enumerated non-sensitive env vars (accessible via dashboard/API), harvested customer credentials. Vercel platform audit logs; behavioral analytics. Moderate. Vercel detected the intrusion after the attacker accessed customer credentials. Detection occurred after exfiltration, not before. Env var access by a compromised Workspace account did not trigger real-time alerting. What’s confirmed vs. what’s claimed Vercel’s bulletin confirms unauthorized access to internal systems, a limited subset of affected customers, and two IOCs tied to Context.ai’s Google Workspace OAuth apps. Rauch confirmed that Next.js, Turbopack, and Vercel’s open-source projects are unaffected. Separately, a threat actor using the ShinyHunters name posted on BreachForums claiming to hold Vercel’s internal database, employee accounts, and GitHub and NPM tokens, with a $2M asking price. Austin Larsen, principal threat analyst at Google Threat Intelligence, assessed the claimant as “likely an imposter.” Actors previously linked to ShinyHunters have denied involvement. None of these claims has been independently verified. Six governance failures the Vercel breach exposed 1. AI tool OAuth scopes go unaudited. Context.ai’s own bulletin states that a Vercel employee granted “Allow All” permissions using a corporate account. Most security teams have no inventory of which AI tools their employees have granted OAuth access to. CrowdStrike CTO Elia Zaitsev put it bluntly at RSAC 2026: “Don’t give an agent access to everything just because you’re lazy. Give it access to only what it needs to get the job done.” Jeff Pollard, VP and principal analyst at Forrester, told Cybersecurity Dive that the attack is a reminder about third-party risk management concerns and AI tool permissions. 2. Environment variable classification is doing real security work. Vercel distinguishes between variables marked “sensitive” (stored in a manner that prevents reading) and those without that designation (accessible in plaintext through the dashboard and API). Attackers used the accessible variables as the escalation path. A developer convenience toggle determined the blast radius. Vercel has since changed its default: new environment variables now default to sensitive. “Modern controls get deployed, but if legacy tokens or keys aren’t retired, the system quietly favors them,” Merritt Baer, CSO at Enkrypt AI and former Deputy CISO at AWS, told VentureBeat. 3. Infostealer-to-SaaS-to-supply-chain escalation chains lack detection coverage. Hudson Rock’s reporting reveals a kill chain that crossed four organizational boundaries. No single detection layer covers that chain. Context.ai’s updated bulletin acknowledged that the scope extended beyond what was initially identified during its CrowdStrike-led investigation. 4. Dwell time between vendor detection and customer notification exceeds attacker timelines. Context.ai detected the AWS compromise in March. Vercel disclosed on Sunday. Every CISO should ask their vendors: what is your contractual notification window after detecting unauthorized access that could affect downstream customers? 5. Third-party AI tools are the new shadow IT. Vercel’s bulletin describes Context.ai as “a small, third-party AI tool.” Grip Security’s March 2026 analysis of 23,000 SaaS environments found a 490% year-over-year increase in AI-related attacks. Vercel is the latest enterprise to learn this the hard way. 6. AI-accelerated attackers compress response timelines. Rauch’s assessment of AI acceleration comes from what his IR team observed. CrowdStrike’s 2026 Global Threat Report puts the baseline at a 29-minute average eCrime breakout time, 65% faster than 2024. Security director action plan Attack Surface What Failed Recommended Action Owner OAuth governance Context.ai held broad “Allow All” Workspace permissions. No approval workflow intercepted. Inventory every AI tool OAuth grant org-wide. Revoke scopes exceeding least privilege. Check both Vercel IOCs now. Identity / IAM Env var classification Variables not marked “sensitive” remained accessible. Accessibility became the escalation path. Default to non-readable. Require a security sign-off to downgrade any variable to accessible. Platform eng + security Infostealer-to-supply-chain Kill chain spanned Lumma Stealer, Context.ai AWS, OAuth tokens, Vercel Workspace, and production environments. Correlate Infostealer intel feeds against employee domains. Automate credential rotation when creds surface in stealer logs. Threat intel + SOC Vendor notification lag Nearly a month between Context.ai detection and Vercel disclosure. Require 72-hour notification clauses in all contracts involving OAuth or identity integration. Third-party risk / legal Shadow AI adoption One employee’s unapproved AI tool became the breach vector for hundreds of orgs. Extend shadow IT discovery to AI agent platforms. Treat unapproved adoption as a security event. Security ops + procurement Lateral movement speed Rauch suspects AI acceleration. Attacker compressed the access-to-escalation window. Cut detection-to-containment SLAs below 29-minute eCrime average. SOC + IR team Run both IoC checks today Search your Google Workspace admin console (Security > API Controls > Manage Third-Party App Access) for two OAuth App IDs. The first is 110671459871-30f1spbu0hptbs60cb4vsmv79i7bbvqj.apps.googleusercontent.com, tied to Context.ai’s Office Suite. The second is 110671459871-f3cq3okebd3jcg1lllmroqejdbka8cqq.apps.googleusercontent.com, tied to Context.ai’s Chrome extension and granting Google Drive read access. If either touched your environment, you are in the blast radius regardless of what Vercel discloses next. What this means for security directors Forget the Vercel brand name for a moment. What happened here is the first major proof case that AI agent OAuth integrations create a breach class that most enterprise security programs cannot detect, scope, or contain. A Roblox cheat download in February led to production infrastructure access in April. Four organizational boundaries, two cloud providers, and one identity perimeter. No zero-day required. For most enterprises, employees have connected AI tools to corporate Google Workspace, Microsoft 365 or Slack instances with broad OAuth scopes — without security teams knowing. The Vercel breach is the case study for what that exposure looks like when an attacker finds it first.