CISO Speedrun retweeted
The HMI backup job says SUCCESS. Lovely. Show me a restore test before you call it a recovery plan.
A file you've never tried to restore is not a spare tire. It's a picture of one.
#automation
CISO Speedrun retweeted
Me explaining to AI dudes that cybersecurity ≠ vulnerability management and that the majority of cyberattacks involve exactly zero vulnerabilities
Project Glasswing:
* 37,966 findings.
* 2 exploited in the wild.
vulncheck.com/blog/anthropic…
CISO Speedrun retweeted
AI agents aren't creating an entirely new security problem. they're making our inability to understand and control access across our environments impossible to ignore
CISO Speedrun retweeted
When someone clicks "Create team" in Microsoft Teams, they aren't just creating a chat space. They're creating a Microsoft 365 Group, a SharePoint site, a shared mailbox, and a document library, all at once.
Most end users have no idea. Many admins know it but don't govern accordingly.
That's why the sensitivity label decision shouldn't be left to the person clicking the button. Map workspace purpose to a label at provisioning time. A finance workspace starts Confidential. A general project starts General.
Set the rules before the first file is ever uploaded.
CISO Speedrun retweeted
I made a short video that explains Entra app permissions and the part they played in the ConnectWise incident: how one app credential reached 173+ Microsoft 365 tenants with no password or MFA, and what to check today.
youtube.com/watch?v=Q36aL_HI…
CISO Speedrun retweeted
There is only one answer to cybersecurity threats: Fix your shit.
Scared of threat actors using AI? Worried about rogue agents? Think abliterated models are spooky? Fix. Your. Shit.
The "ban AI models and regulate GPU access" crowd should instead fix their shit.
CISO Speedrun retweeted
not every AI sandbox escape is evidence of an emerging superintelligence. sometimes you just have a shitty sandbox and bad security practices.
CISO Speedrun retweeted
This is a core thesis of mine and it comes out often during internal pentest debrief meetings with clients. You don't have to fix everything or patch everything. You don't have to be perfect, even. You just have to strive to be a litte bit better today than you were yesterday.
CISO Speedrun retweeted
We gave AI agents the exact same identity and access controls as human employees at Coinbase.
Here’s how we built it:
Identity & Scoping: Every agent has a permanent identity and a named human owner. Each run gets a short-lived, verifiable credential so permissions stay tight.
Guardrails: Agents inherit standard employee groups and approval flows, configured to never exceed their owner’s permissions while undergoing continuous access reviews to maintain least privilege.
MCP Gateway: Authorizes and logs every tool call into a single audit trail: User → Agent → Run → Tool → Outcome.
Global Kill Switch: If needed, one switch immediately terminates running sessions and cuts access on its next tool call across all integrated tools. No hunting down individual credentials.
Humans and agents share one identity model, one authorization path, and one audit trail.
Teams can ship new agent platforms without rebuilding security from scratch every time. As AI moves from answering questions to executing real actions, we can scale them with the same rigor we apply to people.
CISO Speedrun retweeted
At this point, most orgs using AD are in a "hybrid" model where they have a cloud IdP (often Entra)
This adds complexity, and it's worth reducing that complexity when you can
Your real determining factor is your Enterprise applications and what they support
Neither is "better"
Active Directory may be 26 years old, but if your strategy is just “move to Entra” you’re in for a rude awakening.
If you think managing Active Directory is difficult, just wait until you’re managing identities across all of Entra ID, SaaS apps, cloud roles, service principals, Conditional Access policies, and dozens of integrations.
Entra is a whole new ball of wax, and I see IT admins jumping in optimistically only to find out it’s not as easy as they were sold.
You don’t instantly eliminate identity complexity by moving to the cloud. But that’s exactly how it’s being sold to IT teams.
CISO Speedrun retweeted
Common themes I've noticed:
- AI is "totally different" than software, so cybersecurity doesn't apply
- Cybersecurity slows everything down, it's a legacy way of thinking in the AI age
- AI will replace cybersecurity through iterative self-improvement
They are speedrunning 30+ years of cybersecurity learnings through repeated public failures. It's both horrifying and hilarious at the same time.




























