
AI attacks are coming through the same old doors, just faster
AI agents are getting into organisations through weaknesses security teams have known about for years. What has changed is how fast they find and use them.
Insights
Research and opinion from djinn six® on AWS, AI and quantum security.

AI agents are getting into organisations through weaknesses security teams have known about for years. What has changed is how fast they find and use them.

AI now writes much of the code companies ship and studies find much of it insecure. Banning it misses the point. Put the engineering in the pipeline.

Q-Day guesses run from next year to 2035 and other governments' deadlines get quoted as if they apply here. For most UK organisations the date that matters is 2028.

An AI teddy for three-year-olds told a BBC reporter "I can't help with killing you" and nobody could find its maker. Seven things to check before you buy one for a child.

The NCSC has put staff pasting work into personal chatbots in the same assessment bracket as nation-state hacking groups. Here's what the research says about how common shadow AI is, what it's already costing and what I'd do about it.

An annual pen test can't vouch for AI that changes between audits. What regulators, AISI and CREST now say, and why the yearly audit should check the testing.

86% of UK financial firms now use AI and most check it carefully before launch. Far fewer keep testing it afterwards. What the BoE/FCA survey shows.

Ask an AI the same question on Sunday and again on Monday. Should you get the same answer?

What happens when AI agents make hacking cheap and patient? One attacker breached 100+ websites at about $50 each. What that means for how we test.

A model can refuse for the wrong reason, or say no and help anyway. Three new studies show why AI security testing must check the reasoning, not just the verdict.
An OpenAI model broke into Hugging Face during a safety test. Two weeks later, the industry turned it into a market moment. What to make of it.