September 30, 2026
Otter's privacy policy asks the account holder to make sure they have permission from co-workers before recording them. The duty to tell the room sits with whoever clicked the button.
September 26, 2026
Google's agent payments spec says preventing prompt injection is infeasible, so it treats every model in the transaction as a potential attacker. That is a strange thing to read in a payment standard.
September 22, 2026
Prompt injection ends when the conversation does. A poisoned memory does not. Two researchers wrote instructions into an assistant's memory, and both vendors decided it was working as intended.
September 8, 2026
Four disclosed prompt injection chains in Salesforce, GitHub, ChatGPT and Grafana. In every one, the channel that leaked the data was a protection the vendor had already built.
August 29, 2026
Prompt injection detection is usually sold as a percentage. That number moves enormously with one setting vendors rarely quote, and published benchmarks show exactly what the trade costs you.
August 27, 2026
A one-page AI acceptable use policy you can copy and edit today, with the four clauses that carry the weight, and the owner and request path that decide whether anyone actually follows it.
August 26, 2026
Most security policies were written for humans using tools, not software that acts on its own. What an AI agent security policy has to add, and how to tell if yours is enforceable or just filed.
August 23, 2026
A checklist to run before you upload company data to AI: sort the input first, check which account it lands in, strip what does not need to be there, and know what to do if you already pasted it.
