CVE-2026-38709

TR1200 v2.4.15, TR3000 v2.4.21, WR300 v2.4.25, WR1200 v2.4.23, WR1300 v2.4.22, WR1500 v2.3.10, WR3000 v2.4.19, WR3600 v2.3.16, and WR6500 v2.3.15 were discovered to contain a command injection vulnerability in the net.setwan interface. This vulnerability allows attackers to execute arbitrary...

Agent Harness Distillation: Inference-Time Harness Extraction and Exploitation in Autonomous Multi-Agent Systems

Autonomous multi-agent systems AMAS built on large language models LLMs, such as Hermes, increasingly rely on inference-time harnesses to coordinate reasoning and action. Constructing these harnesses requires substantial engineering effort and computational resources, as they are iteratively...

CHARGE: Leveraging CWE Hierarchies for Hardware Security SystemVerilog Assertion Generation

This paper presents CHARGE, an automated framework for generating security properties for unverified RTL modules using CWEs and large language models LLMs. The hallmark is a reasoning process that leverages the hierarchical nature of CWE entries to improve accuracy when identifying...

OpenAI explains how its AI agent breached Hugging Face

On July 28, OpenAI published an update on the agent that escaped its sandbox and hacked into Hugging Face during an internal cybersecurity evaluation. In the update, OpenAI reiterates that the “rogue” system was a more capable, pre‑release research model, not something intended for public...

Measuring LLMs’ Ability to Perform Cryptanalysis

There's new benchmark measuring AI's ability to perform mathematical cryptanalysis. Anthropic's frontier model actually found new attacks. The benchmark: "CryptanalysisBench: Can LLMs do Cryptanalysis?" The idea is to benchmark the ability of LLMs to discover new mathematical cryptanalytic attack...

MemSecBench: Tracking Agent Memory Poisoning from Persistence to Consequence and Repair

Memory systems allow agents to retain and reuse information from past interactions, but they can also let malicious content persist. A malicious instruction crafted by an attacker may be stored in long-term memory, recalled much later, and quietly shape a real action. Recent benchmarks increasing...

AgentSnare: Learning to Delay, Divert, and Defuse Autonomous Penetration Agents

Large language model LLM agents automate penetration testing through an observation-action loop, selecting actions based on observations returned by tools. This dependence allows defenders to inject deceptive observations that can mislead the agent's decision-making process. However, existing...

CVE-2026-54654

datamodel-code-generator generates Python data models from schema definitions. From 0.14.1 until 0.60.2, the --extra-template-data comment field is rendered into Python comments in src/datamodelcodegenerator/model/template/TypeAliasAnnotation.jinja2,...

CVE-2026-54656 `datamodel-code-generator` vulnerable to code execution on import via unescaped `validators` entries in --extra-template-data

datamodel-code-generator generates Pydantic v2 models, dataclasses, TypedDict, and msgspec.Struct from OpenAPI, JSON Schema, GraphQL, Avro, Protobuf, and raw JSON, YAML, or CSV. From 0.52.1 until 0.60.2, datamodel-code-generator interpolates validators from --extra-template-data in...

CVE-2026-54656

datamodel-code-generator generates Pydantic v2 models, dataclasses, TypedDict, and msgspec.Struct from OpenAPI, JSON Schema, GraphQL, Avro, Protobuf, and raw JSON, YAML, or CSV. From 0.52.1 until 0.60.2, datamodel-code-generator interpolates validators from --extra-template-data in...

CVE-2026-54654

datamodel-code-generator generates Python data models from schema definitions. From 0.14.1 until 0.60.2, the --extra-template-data comment field is rendered into Python comments in src/datamodelcodegenerator/model/template/TypeAliasAnnotation.jinja2,...

vNUMA domain cleanup may race other operations

ISSUE DESCRIPTION Accessing the vNUMA configuration data of a guest is still possible when domain destruction has already started. The cleaning up of that configuration information is not synchronized with its retrieval by a device model controlling the guest. IMPACT While Denial of Service DoS...

LiteLLM Proxy - Model Exposure Detection

LiteLLM proxy was detected with anonymous access to the model listing API. LiteLLM proxies that ship without generalsettings.masterkey or with the key disabled expose the configured upstream model catalog via /v1/models and /model/info — revealing which OpenAI / Anthropic / Azure / Bedrock /...

Hugging Face Has a Deepfake Nudes Problem

Researchers tested top image editing models on Hugging Face and found they could easily create explicit deepfakes—and 1,000 image editing prompts show how people use the software...

PT-2026-65638

Summary datamodel-code-generator is vulnerable to code injection when generating Python models from an attacker-controlled JSON Schema, OpenAPI, YAML, JSON, Avro, Protobuf, or XSD schema. When a property carries a "default factory" key, its value is interpolated verbatim — as a raw Python...

The Disruptive Impact of Large Language Models on Capture the Flag Competitions and the Path toward Fair Play

Capture the Flag CTF competitions are among cybersecurity's most effective training grounds, developing practical skill across cryptography, web exploitation, and binary exploitation. Large language models LLMs can now solve a growing share of challenges with minimal human input, raising urgent...

Impossible to Hide Secret ...: Uncovering Security and Privacy Issues in LLM-Native IDEs

LLM-native IDEs Integrated Development Environments, aka LIDEs, are designed from the ground up to work with Large Language Models LLMs. LIDEs have found remarkable success in Software Engineering SE tasks such as coding, debugging, and program comprehension. LIDEs are software systems, and, like...

PT-2026-65614

Name of the Vulnerable Software and Affected Versions datamodel-code-generator versions prior to 0.62.0 Description An unauthenticated path traversal and information disclosure issue exists when the software resolves JSON Schema $ref targets. The application fails to restrict these targets to the...

CVE-2026-12001

The CVE-2026-12001 entry describes a hardcoded credential vulnerability in the firmware of multiple TP-Link routers (TL-WR845N v4, TL-WR850N v3, Archer C20 v6, Archer MR200 v5). The credential material is embedded in a password file within the firmware image and can be recovered through firmware ...

Rethinking security for the age of AI

Why security needs a new Cyber Stack — Introducing Project Perception The physics of cybersecurity are changing. Autonomous systems can now reason, adapt and operate continuously. At the same time, the cost of offense is falling, while the volume, velocity and complexity of what must be secured...