| Titel | VectifyAI PageIndex latest Infinite Loop (CWE-835) |
|---|
| Beschreibung | # Technical Details
An Arbitrary Infinite Loop Denial of Service (DoS) vulnerability exists in the `toc_transformer` method in `pageindex/page_index.py` of PageIndex.
The application fails to limit loop iterations when handling LLM responses that hit the maximum output token limit (where `finish_reason == "length"`). Without a retry counter, the system automatically and endlessly attempts to instruct the model to "continue".
# Vulnerable Code
File: `pageindex/page_index.py`
Method: `toc_transformer()`
Why: The while loop rebuilds the prompt and calls the LLM API without checking an iteration limit or bailout condition when the completion check repeatedly fails.
# Reproduction
1. Generate an adversarial PDF with thousands of sections in the TOC (sufficiently large to cause the LLM to truncate output), or set up a Mock OpenAI proxy that forcibly returns `finish_reason: "length"`.
2. Run the application via the CLI against the malicious PDF: `python run_pageindex.py --pdf_path evil_toc.pdf --model gpt-3.5-turbo`
3. Observe the process forever attempting to complete the TOC, utilizing 100% of a CPU thread and rapidly emitting API requests.
# Impact
- Denial of Service (DoS) due to process hanging.
- Immediate financial exhaustion of the backend LLM service billing account due to unbounded API requests. |
|---|
| Quelle | ⚠️ https://github.com/VectifyAI/PageIndex/issues/174 |
|---|
| Benutzer | Eric-b (UID 96354) |
|---|
| Einreichung | 23.04.2026 09:40 (vor 1 Monat) |
|---|
| Moderieren | 11.05.2026 15:16 (18 days later) |
|---|
| Status | Akzeptiert |
|---|
| VulDB Eintrag | 362605 [VectifyAI PageIndex bis f50e52975313c6716c02b20a119577a1929decba PDF Table of Contents pageindex/page_index.py toc_transformer Denial of Service] |
|---|
| Punkte | 20 |
|---|