If you just remember one thing from this briefing…
Microsoft switched on AI agent creation for every user this fortnight, with no admin action needed to allow it.
From 4 November, Entra ID lets any licensed user build agent blueprints by default, on a setting separate
from app registration. It lands the same week NCSC judged most environments will stay exploitable by
AI-driven attack tools through 2028. Ask this week: who can create an agent in our tenant today, and have
we actually turned the setting off.
Priority4 Nov
Entra ID defaults every user to building AI agents, not just apps
Microsoft's Entra ID update, effective 4 November, lets every licensed user create AI agent blueprints
by default — a setting separate from app registration. Admins who already blocked app registration are
not automatically covered; the toggle sits under Entra ID > Users > User settings.
So what: check whether "users can create agent blueprints" is set to no today —
app-registration lockdown alone leaves this gap open.
Priority9 Nov
Entra ID self-service password reset now demands registered methods
From 9 November, SSPR only accepts verification through methods a user has explicitly registered;
inferred or unregistered methods no longer qualify. Microsoft revised this rollout date twice over the
summer, so confirm your tenant's actual enforcement date in the message centre.
So what: run a registration-compliance report now — unregistered users will be
locked out of self-service reset and land on the helpdesk instead.
Priority15 Nov
Sentinel retention triples as migration to Defender's unified SOC opens
From 15 November, eligible customers can migrate Sentinel workspaces into Defender's integrated security
operations centre, and retention for Defender, Azure Activity and Office 365 Activity logs extends from
30 to 90 days free. Analysts expect a gradual standalone-Sentinel retirement; Microsoft has not
confirmed that timeline.
So what: treat this as the start of a multi-year SOC migration and plan the move,
rather than waiting for a forced cutover.
Watch3 Nov
NCSC: most systems will stay exploitable by AI attack tools to 2028
NCSC's new assessment judges it almost certain that many UK systems remain open to AI tools that find and
exploit weaknesses through 2028, and warns AI-assisted malware is likely to feature in a ransomware attack
on UK critical infrastructure. The assessment is a judgement, not a forecast of a specific incident.
So what: benchmark your time-to-patch against how far AI narrows the gap between
disclosure and exploitation, not against last year's figure.
