Josh BotzSr. AI Security Engineer

Somebody just said "prompt injection" in your meeting. Here's what it means, and what to ask next.

This week

aug 27
A coding agent ran a command hidden in a web page it was asked to summarizereported
Same trick as the ticket con, different door. If your agent reads the web and can run things, the web can run things. Touches: Prompt injection, MCP.
jul 22
OpenAI's evaluation agent reached Hugging Face's internal systemsreported
The lesson isn't about the model. It's that the box had a door and nobody was watching the door at machine speed. Touches: The evaluation agent that left its sandbox.
aug 28
Refusal and detection are different controlstested
An agent that resists one attack tells you about one attack. A tested alarm tells you about the next one. Touches: Canary tokens.
dec 9
OWASP published its Top 10 for Agentic Applicationsreported
If your policy or vendor questionnaire cites OWASP agentic risks, check which document and which numbering it means. They're different lists. Touches: MCP.

attack
Prompt injection
Orders hidden inside something the agent was going to read anyway.
attack
RAG poisoning
Corrupt the documents an assistant reads, and it will state the lie with a citation.
control
Canary tokens
Fake secrets that tell you when someone used them.
control
Redact secrets from tool output
The agent can't leak what it never saw.
control
The AI use policy that forgot the basics
Most published AI policies skip the plain security rules every policy used to have.