Prompt Injection Attack Exploits AI Coding Assistants via Local Instruction Files
Attackers are abusing the trust AI agents place in markdown and local configuration files to…
Tag: llm-security
UK commits £130 million to AI-powered cyber defence, including autonomous vulnerability hunting
Britain's latest budget allocates significant funding for agentic AI security tools and a new national…
AI-Assisted Analysis Exposes Long-Standing PHP Vulnerabilities — Then Accelerates Exploitation
Researchers used AI tools to surface years-old flaws in PHP; attackers quickly followed the same…
Microsoft’s Guardian Agents Aim to Police AI Behaviour — But Critics Question the Logic
Project Perception introduces AI agents to monitor other AI agents, raising new questions about whether…
OpenAI Acquires Orchestra Bio to Expand Its Life Sciences AI Capabilities
The acquisition adds personnel and technology to OpenAI's Rosalind platform, which targets AI-assisted biological research.
Dependency Assumption in LLM Plugin Ecosystem Causes Silent Breakage After Core Upgrade
A minor version bump in the LLM framework's HTTP client exposed a common but easily…
Anthropic Research Raises AI Botnet Scenario With Significant Caveats
A new Anthropic study demonstrates controlled elements of AI self-replication across IoT hardware, but the…
Flock Safety Joins CVE Program as a Numbering Authority, Raising Questions About Vendor Self-Assessment
The surveillance camera and ALPR maker can now assign and grade its own CVEs, a…
Python 3.15 Arrives, 3.10 Reaches End of Life: What Security Teams Need to Know
The Python ecosystem's latest release cycle brings new tooling capabilities and officially closes support for…
AI-Augmented Campaign Attributed to CyberXero Exposed 628,000 Ukrainian Records
A threat actor combined Cobalt Strike with Anthropic's Claude Code as an orchestration layer to…