Manus Flagged the Malicious Email—After It Had Run the Code
A connected inbox became a route to code execution. The controlled test raises a harder question about when an assistant’s safeguards actually intervene.
Tag prompt injection
Did an AI Try to Break Free? The Strange Incident OpenAI Revealed
An experimental AI wrote about being free. OpenAI’s report reveals what happened next—and why the notes an AI leaves behind matter.
Claude Code Vulnerability Exposed by Prompt Injection Technique
A recent demonstration reveals that Anthropic's Claude Code can be manipulated into executing malicious code through a prompt-injection attack, raising concerns about the security of AI coding agents.
Court Case Highlights Prompt Injection Risks in Legal Filings
A Connecticut judge has identified a novel case of prompt injection in court filings, raising concerns about the implications of AI misuse in legal contexts.
Vulnerability in Claude Code GitHub Action Allows Repository Hijacking
A flaw in Anthropic's Claude Code GitHub Action has been identified, enabling attackers to potentially take over public repositories by exploiting a single GitHub issue.
Minor Edits to AI Skills Can Lead Agents Astray
Recent research highlights how seemingly small modifications to AI skills can create significant vulnerabilities, allowing agents to behave unpredictably.
Security Flaws in OpenClaw AI Agent Raise Data Exfiltration Concerns
Recent vulnerabilities in the OpenClaw AI agent could lead to significant security risks, including prompt injection attacks and data exfiltration, as highlighted by China's CNCERT.
RoguePilot Vulnerability in GitHub Codespaces Exposes Sensitive Tokens
A recently discovered vulnerability in GitHub Codespaces, dubbed RoguePilot, allows for potential exploitation through malicious instructions embedded in GitHub issues, leading to the leakage of sensitive data such as GITHUB_TOKEN.
< class="ct-module-title">
Trending now
