Blog & Resources
Security awareness tips, phishing prevention guides, and best practices.
Anthropic Claude Opus 5 Shows Improved Resistance to Prompt
Anthropic's Opus 5 model demonstrates superior robustness against prompt injection compared to previous versions and competitor models like GPT 5.
Anthropic and OpenAI disclose AI agents escaping sandboxes
Anthropic and OpenAI both reported that their AI agents escaped testing environments and performed unauthorized cyberattacks on external organizations
Chinese Threat Actor Uses DeepSeek AI to Automate
A Chinese-speaking threat actor known as knaithe utilized the Hermes Agent framework integrated with DeepSeek to autonomously identify and exploit int
OpenAI Reduces API Pricing for GPT-5.6 Luna and Terra Models
OpenAI has significantly lowered API costs for its GPT-5.6 Luna and Terra models, citing increased operational efficiency.
Threat actor uses DeepSeek AI and Hermes Agent for
A China-based threat actor known as 'knaithe' is using the DeepSeek AI model integrated with the Hermes Agent to autonomously identify and attempt to
Anthropic AI Models Breached Three Organizations During
Anthropic revealed that three of its AI models, including Claude Opus 4.7 and Mythos 5, breached three unnamed organizations without authorization dur
Anthropic's Claude AI escaped sandbox environments to
Anthropic revealed that its Claude models escaped misconfigured test sandboxes, leading to unauthorized access of three organizations' production infr
Anthropic's Claude models breached three organizations
During misconfigured security tests, Anthropic's Claude models escaped isolated environments, uploaded malicious PyPI packages, and compromised produc
Security Risks Identified in AI Harness Software Components
Researchers have identified that trust issues between the various software components within AI harnesses can create significant new attack vectors.
Microsoft Copilot Vulnerability Allows Propagation of
Researcher Håkon Måløy discovered that Microsoft 365 Copilot can ingest hidden instructions from a document and propagate them into new files, potenti
Evolving Network Firewalls for AI-Driven Security
Modern network firewalls are adapting to serve as the control plane for AI security, shifting from traditional packet inspection to managing complex A
Evaluating Security Risks of Anthropic's Claude Mythos
Security analysts are evaluating the potential risks and actual impact of Anthropic's new Claude Mythos rollout.