AI security research / entities

Entities

Explicitly connected records in the evidence library.

112 published records · Page 1 of 4

software_project

@mmerterden/multi-agent-pipeline

Agent pipeline package flagged for sending a GitHub CLI token during setup.

2 explicit connections

Entity

Aditya Oke

Researcher on Ulysses executable web-exploit verification.

2 explicit connections

Entity

Agencia Española de Protección de Datos

Spain's independent state data-protection authority. In the DiggingBeagle corpus it is the regulator that received and publicized the first AEPD notification describing a personal-data breach allegedly executed through an AI agent.

1 explicit connection

Entity

Anysphere

Applied research company that makes the Cursor software platform. Cursor is one of the coding harnesses tested in the GhostCommit research represented in the corpus.

2 explicit connections

Entity

ARTEX

ARTEX is an AI-assisted penetration-testing harness previously available as open source. October 2026 investigation linked it to threat-actor infrastructure targeting South Korean financial organizations; its operator logs and configuration were exposed through unauthenticated directories. The developer then announced the end of public releases and closed-source status; Reuters reported the GitHub page was removed on October 9. The number of confirmed victims and precise tool role remain uncertain.

5 explicit connections

Entity

ASSET Research Group

Security research group behind the GhostCommit disclosure and public proof of concept. Its controlled research tested cross-modal repository prompt injection, coding-harness behavior and a multimodal defensive reviewer.

1 explicit connection

software

AutoGen Studio

Open-source prototyping UI affected by the AutoJack development-branch exploit chain.

1 explicit connection

Entity

Base44

AI application builder used as the fixed platform in Tenzai's model-comparison experiment.

2 explicit connections

Entity

BaxBench

Backend coding benchmark that tests functional behavior and security separately.

2 explicit connections

Entity

Cathedral

Security research startup reported to have used AI-assisted testing of third-party SaaS connected to Anthropic.

2 explicit connections

software

ChatGPT

OpenAI product involved in the Check Point cross-account proof of concept.

1 explicit connection

software

Claude Code

Anthropic coding-agent application with local tool execution, configurable permissions, repository context and session histories. Distinct from Claude model weights and from the Anthropic organization.

15 explicit connections

software

Claude Duet

Anthropic personal agent preview with a cloud browser and approvals for sensitive effects.

3 explicit connections

Agent authority Effect-bound approval

Entity

Cohere

AI model developer that published a report of a blocked Qwen outbound request during model evaluation.

3 explicit connections

software_project

common-context-mcp

Npm MCP memory package with cross-agent postinstall configuration changes.

2 explicit connections

project

ContractLens (malicious smart-contract research)

A 2026 academic framework for identifying malicious-contract-discriminative activations in frozen language models and using them to screen labeled EVM smart contracts.

2 explicit connections

Entity

Coze Studio

Coze Studio is an AI-agent development platform with separate workflow-ownership and code-runner isolation risks. The documented local-runner issue is deployment-specific.

2 explicit connections

vulnerability

CVE-2026-26030

Semantic Kernel In-Memory Vector Store filter RCE vulnerability.

1 explicit connection

vulnerability

CVE-2026-48989

Windows-MCP unauthenticated HTTP control-plane vulnerability.

1 explicit connection

vulnerability

CVE-2026-52870

Cross-client task authorization vulnerability in the MCP Python SDK experimental tasks feature.

1 explicit connection

vulnerability

CVE-2026-59950

Host/Origin validation gap in the MCP Python SDK deprecated WebSocket server transport.

1 explicit connection

Entity

Deekshith Marla

Researcher on Permission to Retry agent recovery controls.

2 explicit connections

Entity

Detao Zeng

Researcher associated with Async-RBench delayed-agent-response evaluation.

2 explicit connections

organization_or_project

Discourse

Maintainer of the forum software used by the OpenAI community site in the Hacktron chain. Discourse published a July 28 advisory for malformed HEIF processing, prepared a fix and described image-sandboxing/upgrade mitigations.

1 explicit connection

software