Steering LLM Viewpoints through Fabricated Evidence Injection

As chatbots increasingly influence daily decision-making, their potential to produce misleading responses poses substantial risks to users. This paper investigates a critical cognitive vulnerability in LLMs: their tendency to uncritically trust external context when presented with fabricated...

pytorch-vision-classifier (>=0.0.4, <=0.0.8), train-and-deploy (>=0.1.1a0, <=0.1.4) +2 more potentially affected by CVE-2026-4035 via mlflow (1.10.0)

mlflow PYPI version 1.10.0 is affected by a known vulnerability. The following packages have a transitive dependency on mlflow and may be impacted: - pytorch-vision-classifier =0.0.4, =0.1.1a0, =0.4.1, =0.4.2 - waveletai 0.2.4 Source cves: CVE-2026-4035 Source advisory: OSV:GHSA-G35P-PX32-WHV6...

pytorch-vision-classifier (>=0.0.4, <=0.0.8), train-and-deploy (>=0.1.1a0, <=0.1.4) +2 more potentially affected by CVE-2026-3198 via mlflow (1.10.0)

mlflow PYPI version 1.10.0 is affected by a known vulnerability. The following packages have a transitive dependency on mlflow and may be impacted: - pytorch-vision-classifier =0.0.4, =0.1.1a0, =0.4.1, =0.4.2 - waveletai 0.2.4 Source cves: CVE-2026-3198 Source advisory: OSV:GHSA-R5M9-WM49-959F...

arvados-cwl-runner (>=2.6.0rc3, <=2.7.4), db-classifier (0.0.2) +18 more potentially affected by CVE-2026-44897 via mistune (2.0.5)

mistune PYPI version 2.0.5 is affected by a known vulnerability. The following packages have a transitive dependency on mistune and may be impacted: - arvados-cwl-runner =2.6.0rc3, =2.0.5, =4.1.0, =0.8.4, =2.1.1, =1.0.0, =2.2.0, =0.0.1, =2.4.1rc3, =8.3.20221209165047, =8.5.20240102191335 - sitefa...

arvados-cwl-runner (>=2.6.0rc3, <=2.7.4), db-classifier (0.0.2) +18 more potentially affected by CVE-2026-44708 via mistune (2.0.5)

mistune PYPI version 2.0.5 is affected by a known vulnerability. The following packages have a transitive dependency on mistune and may be impacted: - arvados-cwl-runner =2.6.0rc3, =2.0.5, =4.1.0, =0.8.4, =2.1.1, =1.0.0, =2.2.0, =0.0.1, =2.4.1rc3, =8.3.20221209165047, =8.5.20240102191335 - sitefa...

pytorch-vision-classifier (>=0.0.4, <=0.0.8), train-and-deploy (>=0.1.1a0, <=0.1.4) +2 more potentially affected by CVE-2026-2651 via mlflow (1.10.0)

mlflow PYPI version 1.10.0 is affected by a known vulnerability. The following packages have a transitive dependency on mlflow and may be impacted: - pytorch-vision-classifier =0.0.4, =0.1.1a0, =0.4.1, =0.4.2 - waveletai 0.2.4 Source cves: CVE-2026-2651 Source advisory: OSV:GHSA-8C7Q-86FQ-VVMH...

Disentangling Adversarial Prompts: A Semantic-Graph Defense for Robust LLM Security

Large Language Models LLMs are increasingly vulnerable to adversarial prompts that exploit semantic ambiguities to bypass safety mechanisms, resulting in harmful or inappropriate outputs. Such attacks, including jailbreaking and prompt injection, pose significant risks to the integrity and...

decoders (>=0.0.7, <=0.0.12), fast-token-classifier (>=0.1.0, <=0.2.3) +1 more potentially affected by CVE-2026-4372 via transformers (4.32.0)

transformers PYPI version 4.32.0 is affected by a known vulnerability. The following packages have a transitive dependency on transformers and may be impacted: - decoders =0.0.7, =0.1.0, =0.2.3 - myllm 4.4.0 Source cves: CVE-2026-4372 Source advisory: OSV:PYSEC-2026-2289...

esmerald (0.4.0), fast-token-classifier (>=0.1.0, <=0.2.3) +2 more potentially affected by CVE-2026-48710 via starlette (0.23.0)

starlette PYPI version 0.23.0 is affected by a known vulnerability. The following packages have a transitive dependency on starlette and may be impacted: - esmerald 0.4.0 - fast-token-classifier =0.1.0, =0.2.3 - fastapi 0.90.0 - fastapi-stack-utils 0.8.4 Source cves: CVE-2026-48710 Source advisor...

Blind Spots in the Guard: How Domain-Camouflaged Injection Attacks Evade Detection in Multi-Agent LLM Systems

Injection detectors deployed to protect LLM agents are calibrated on static, template-based payloads that announce themselves as override directives. We identify a systematic blind spot: when payloads are generated to mimic the domain vocabulary and authority structures of the target document, wh...

pytorch-vision-classifier (>=0.0.4, <=0.0.8), train-and-deploy (>=0.1.1a0, <=0.1.4) +2 more potentially affected by CVE-2026-2652 via mlflow (1.10.0)

mlflow PYPI version 1.10.0 is affected by a known vulnerability. The following packages have a transitive dependency on mlflow and may be impacted: - pytorch-vision-classifier =0.0.4, =0.1.1a0, =0.4.1, =0.4.2 - waveletai 0.2.4 Source cves: CVE-2026-2652 Source advisory: OSV:GHSA-75CM-X2W3-8MGF...

pytorch-vision-classifier (>=0.0.4, <=0.0.8), train-and-deploy (>=0.1.1a0, <=0.1.4) +2 more potentially affected by CVE-2026-2652 via mlflow (1.10.0)

mlflow PYPI version 1.10.0 is affected by a known vulnerability. The following packages have a transitive dependency on mlflow and may be impacted: - pytorch-vision-classifier =0.0.4, =0.1.1a0, =0.4.1, =0.4.2 - waveletai 0.2.4 Source cves: CVE-2026-2652 Source advisory: OSV:PYSEC-2026-2221...

Snorkel MultitaskClassifier.load uses an unsafe torch.load

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...

GHSA-GPX5-7XM4-229W Snorkel MultitaskClassifier.load uses an unsafe torch.load

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...

Deserialization of Untrusted Data

Overview snorkel is an A system for quickly generating training data with weak supervision Affected versions of this package are vulnerable to Deserialization of Untrusted Data in the MultitaskClassifier.load function. An attacker can execute arbitrary code by supplying a maliciously crafted mode...

EUVD-2026-29508

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...

CVE-2026-31224

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...

CVE-2026-31224

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...

CVE-2026-31224

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...

CVE-2026-31224

The snorkel library thru v0.10.0 contains an insecure deserialization vulnerability CWE-502 in the MultitaskClassifier.load method of the MultitaskClassifier class. The method loads model weight files using torch.load without enabling the security-restrictive weightsonly=True parameter. This...