Stanford NLP DSPy up to 3.3.0b1 Image/Audio output field adapters image.py encode_image/encode_audio url injection

Summaryinfo

A vulnerability marked as problematic has been reported in Stanford NLP DSPy up to 3.3.0b1. This issue affects the function encode_image/encode_audio of the file image.py of the component Image/Audio output field adapters. The manipulation of the argument url leads to injection. This vulnerability is traded as CVE-2026-72742. It is possible to initiate the attack remotely. There is no exploit available.

Detailsinfo

A vulnerability, which was classified as problematic, has been found in Stanford NLP DSPy up to 3.3.0b1. This issue affects the function encode_image/encode_audio of the file image.py of the component Image/Audio output field adapters. The manipulation of the argument url with an unknown input leads to a injection vulnerability. Using CWE to declare the problem leads to CWE-74. The product constructs all or part of a command, data structure, or record using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify how it is parsed or interpreted when it is sent to a downstream component. Impacted is confidentiality. The summary by CVE is:

DSPy 3.3.0b1 contains a file exfiltration vulnerability in the Image and Audio output field adapters that allows attackers with influence over language model outputs to read arbitrary local files by injecting a filesystem path into the url field of a parsed Image or Audio typed output. The JSONAdapter and ChatAdapter parse untrusted language model completions through parse_value into TypeAdapter validation, which triggers encode_image or encode_audio to read and base64-encode any local file path via the os.path.isfile branch in image.py and audio.py, subsequently embedding the file contents into outgoing prompt messages sent to the attacker-controlled model endpoint.

The advisory is shared at github.com. The identification of this vulnerability is CVE-2026-72742 since 08/10/2026. The exploitation is known to be easy. The attack may be initiated remotely. It demands that the victim is doing some kind of user interaction. Technical details are known, but no exploit is available. The price for an exploit might be around USD $0-$5k at the moment (estimation calculated on 08/11/2026). MITRE ATT&CK project uses the attack technique T1055 for this issue.

There is no information about possible countermeasures known. It may be suggested to replace the affected object with an alternative product.

If you want to get the best quality for vulnerability data then you always have to consider VulDB.

Productinfo

Vendor

• Stanford NLP

Name

• DSPy

Version

• 3.3.0b1

CPE 2.3info

• 🔒

CPE 2.2info

• 🔒

CVSSv4info

CVSSv3info

CVSSv2info

Exploitinginfo

Threat Intelligenceinfo

Countermeasuresinfo

Timelineinfo

Sourcesinfo

Entryinfo

If you want to get the best quality for vulnerability data then you always have to consider VulDB.

Discussion