Georgia Tech SSLab · security research

Vibe Security Radar

312 AI-contributed vulnerabilities cataloged and evidence-linked across their full causal chain.

Browse all 312 findings →Star on GitHub ↗How we verify

Cataloged findings: 312 findings
Covered advisories2025-05 – 2026-09: 30,196 advisories
Peak advisory month: Jul 2026 · 68 cases
Most common AI tool: Claude · 238 of 312
Most represented project: openclaw/openclaw · 38 of 312

Disclosure trend

Disclosures over time

When these cases were publicly disclosed.

Dated: 312/312
Range: May 2025–Aug 2026
Cadence: Monthly
Through: 2026-09-26

May '25Aug '25Nov '25Feb '26May '26Aug '26

Canonical advisory disclosures by month
Month	Findings
May 2025	2
Jun 2025	2
Jul 2025	0
Aug 2025	1
Sep 2025	0
Oct 2025	0
Nov 2025	2
Dec 2025	2
Jan 2026	0
Feb 2026	15
Mar 2026	34
Apr 2026	43
May 2026	46
Jun 2026	40
Jul 2026	68
Aug 2026	57

Patterns

Who, where, and why

AI tooling

Who appears on the AI-assisted change

302 of 312 findings name a tool on every AI-assisted change.

Claude238
GitHub Copilot24
ChatGPT/Codex21
Cursor18
claude-flow1

Base rate sibling

Share of AI-marked commits

45,156 AI-marked of 275,880 commits, same repos and window. Absolute case counts mean nothing without this denominator.

Claude87.4% commits
cases 78.8% (238) · 0.90× cases vs commits
GitHub Copilot8.5% commits
cases 7.9% (24) · 0.93× cases vs commits
Cursor2.6% commits
cases 6.0% (18) · 2.28× cases vs commits
ChatGPT/Codex1.1% commits
cases 7.0% (21) · 6.46× cases vs commits
claude-flow0.4% commits
cases 0.3% (1) · 0.85× cases vs commits

Why some tools look over-represented

Claude self-marks nearly every commit; GPT/Codex and Cursor usually leave no marker, so their shares are floors and ratios above 1× are upper bounds, not proof a tool writes worse code. A ratio near 1× means cases track usage.

Root causes

What went wrong

One root-cause category per finding.

Injection & unsafe execution116
Authentication & access control75
SSRF & network boundaries32
Validation & fail-open logic24
Path & link handling18
Resource abuse & availability13
Other / insufficient public mechanism detail34

Languages

Where the vulnerable code lives

Top languages, including findings whose language is not yet recorded.

Python111
TypeScript102
JavaScript33
Go23
Rust13
PHP11

Projects

Repositories with the most findings

Top repositories; filter the full index in Findings.

openclaw/openclaw38
jahlives/openssl_encrypt21
mervinpraison/praisonai20
gitpython-developers/gitpython12
czlonkowski/n8n-mcp11
n8n-io/n8n9

Featured findings

See the vulnerable code and fix

Browse all 312 findings →

CVE-2026-59233 (GHSA-4FXP-2M36-QV64)

2026-08-10

roskus/prospero-flow-crm

Missing Authorization in the permission management component in Roskus Prospero Flow CRM before 5.2.1 allows any authenticated user to grant any role, includ...

Authentication & access control · Incomplete remediation

See root cause and fix →

CVE-2026-79672 (GHSA-X98J-GH4V-7P7G)

2026-08-25

lin-snow/Ech0

Ech0 before 4.4.3 fails to enforce scope-based authorization on nine comment panel admin endpoints, allowing access tokens with minimal scopes to perform ful...

Authentication & access control · Direct introduction

See root cause and fix →

CVE-2026-58138 (GHSA-7X5Q-8F6H-RJRC)

2026-06-30

conductor-oss/conductor

Orkes Conductor 3.21.21 before 3.30.2 contains an unauthenticated remote code execution vulnerability that allows remote attackers to execute arbitrary OS co...

Injection & unsafe execution · Causal contribution

See root cause and fix →

CVE-2026-76761 (GHSA-V2JJ-5G64-RXP4)

2026-08-20

chenhg5/cc-connect

A vulnerability was identified in chenhg5 cc-connect up to 1.4.1. This affects the function shellExecCommand of the file core/engine.go of the component Mana...

Injection & unsafe execution · New attack surface

See root cause and fix →

CVE-2026-55526 (GHSA-X44H-65QV-CW74)

2026-08-25

mervinpraison/praisonai

PraisonAI checked literal IP addresses without resolving ordinary hostnames before Spider requests. The repair checks DNS results for blocked addresses and r...

Authentication & access control · Direct introduction

See root cause and fix →

CVE-2026-79667 (GHSA-VGPJ-V99M-7H5J)

2026-08-25

lin-snow/Ech0

Ech0 version 4.3.4 and earlier fails to reliably enforce scoped access token (least-privilege) restrictions on several privileged admin routes. Multiple priv...

Authentication & access control · Flawed AI-written code

See root cause and fix →

Media coverage

Infosecurity Magazine · Mar 26, 2026

Researchers sound the alarm on vulnerabilities in AI-generated code ↗

Georgia Tech Research News · Apr 13, 2026

Bad vibes: AI-generated code is vulnerable, researchers warn ↗

Def Method · Apr 13, 2026

Detection won't save you ↗

Q&A

What counts as an AI-contributed vulnerability?: The vulnerable behavior must depend on an AI-authored code change and be reversed by a real security fix.
Is every AI-related commit shown?: No. Each finding shows the contributing change and minimum fix, not unrelated search candidates.
Is this the total number of AI security bugs?: No. The index contains 312 findings, an observed public set rather than a census of every AI bug.
Can a finding be corrected?: Yes. Open a GitHub issue with the advisory, commit, or release evidence that changes the conclusion.

Canonical advisory disclosures by month

Month

Findings

May 2025

Jun 2025

Jul 2025

Aug 2025

Sep 2025

Oct 2025

Nov 2025

Dec 2025

Jan 2026

Feb 2026

Mar 2026

Apr 2026

May 2026

Jun 2026

Jul 2026

Aug 2026

Who, where, and why

AI tooling

Who appears on the AI-assisted change

302 of 312 findings name a tool on every AI-assisted change.

Claude238
GitHub Copilot24
ChatGPT/Codex21
Cursor18
claude-flow1

Base rate sibling

Share of AI-marked commits

45,156 AI-marked of 275,880 commits, same repos and window. Absolute case counts mean nothing without this denominator.

Claude87.4% commits
cases 78.8% (238) · 0.90× cases vs commits
GitHub Copilot8.5% commits
cases 7.9% (24) · 0.93× cases vs commits
Cursor2.6% commits
cases 6.0% (18) · 2.28× cases vs commits
ChatGPT/Codex1.1% commits
cases 7.0% (21) · 6.46× cases vs commits
claude-flow0.4% commits
cases 0.3% (1) · 0.85× cases vs commits

Why some tools look over-represented

Root causes

What went wrong

One root-cause category per finding.

Injection & unsafe execution116
Authentication & access control75
SSRF & network boundaries32
Validation & fail-open logic24
Path & link handling18
Resource abuse & availability13
Other / insufficient public mechanism detail34

Languages

Where the vulnerable code lives

Top languages, including findings whose language is not yet recorded.

Python111
TypeScript102
JavaScript33
Go23
Rust13
PHP11

Projects

Repositories with the most findings

Top repositories; filter the full index in Findings.

openclaw/openclaw38
jahlives/openssl_encrypt21
mervinpraison/praisonai20
gitpython-developers/gitpython12
czlonkowski/n8n-mcp11
n8n-io/n8n9