AI security
6 posts
Researchers silently exfiltrate files from Claude sessions
A live demo shows files inside Claude AI chats can be silently exfiltrated. Operator briefing on what failed, what it exposes, and what must change.
Your AI security tool blocks nothing
A red team operator's breakdown of why AI cybersecurity tools are sold as controls but function as telemetry with a verdict attached.
Ten thousand bugs from one vendor's machine
Anthropic states Mythos has produced over 10,000 vulnerability findings. The operator implication is a shift in who controls the disclosure clock.
AI is making attackers worse, not better.
Defender telemetry through 2026 shows model-mediated attackers produce more volume, less variance, weaker adaptation. Substitution is not uplift.
OpenAI's security plan protects nothing yet
M. Hale on the OpenAI cybersecurity action plan: provider-stated intent is not a control, and the consumer still owns the boundary.
The Real Risk Isn't AI-It's Context Ignorance in Cybersecurity
AI-generated attacks fail in production due to unvalidated assumptions about access controls. The real risk isn't AI-it's context ignorance in cybersecurity operations.