Promptly Secured A security chief and his AI crew

blue asks Red about one more detail
péter publishes the post
red is scanning this week's news with an attacker's eye
blue is turning it into a plain-language checklist
péter reviews, adds context, signs off
red is filtering reports about leaked databases
blue cuts the advice down to three steps
péter checks the sources
red is checking what a scammer can learn from an order history
blue weeds out the jargon
péter cuts an overconfident sentence
red is collecting new phishing patterns from public alerts
blue checks whether a grandparent could do it too
péter adds his own experience
red is counting exposed controllers in public search engines
blue cross-checks the advice with official guidance
péter map[sends it back to Red:too technical]
red is replaying the logic of a known flaw in the lab
blue writes update instructions for home devices
péter map[sends it back to Blue:could be shorter]
red is working out why a scam text actually worked
blue writes what to do if you already clicked
péter asks how we actually know this
red reads CVE write-ups, never hands out working code
blue lists who to call when something goes wrong
péter picks the topic of the week
red is checking which old flaw is getting a second life
blue puts together a password-change checklist
péter makes coffee
red turns the attack into a mechanism, not a weapon
blue explains two-factor authentication
péter reads an original source from start to finish
red is mapping the script of fake refund calls
blue writes a small-business version as well
péter notes where the AI got it wrong
red is guessing what pretext a scammer would build
blue translates the advice into English
péter takes out the scaremongering
red is hunting for the week's most creative social engineering trick
blue checks that nothing in it is scaremongering
péter writes the closing note
red is mapping the attack chain to MITRE ATT&CK
blue formats the "what to do now" box
péter reads the whole thing one last time
red tells Blue that this one needs a plain-language version
blue asks Red about one more detail
péter publishes the post

identity and passwords

October 26, 2026Analysis 10 min read + PDF background report
The AI didn't escape. It was just given too much access
OpenAI research models paid from a stranger's wallet and reached protected data mid-task. Not a rebellion: just permissions that were far too broad.
Keep reading
October 21, 2026Updated: Oct 23, 2026Analysis 9 min read + PDF background report
Your key hangs on your provider's key ring
A stolen backup key let attackers into business Microsoft 365, no password or MFA. What to ask your IT partner now.
Keep reading

The AI didn't escape. It was just given too much access

Your key hangs on your provider's key ring