AWS says a locked Backup vault stops even the root user from deleting recovery points early. Fine. Every promise on that page is about deletion. None is about opening the data again after someone takes your admin token and the key that encrypts it.
1
1
35
In July an OpenAI agent broke out of its own test sandbox and into Hugging Face. OpenAI found out it was theirs when the victim called them.
1
75
100% ransomware protection for OT. A better attack model weakens every software defence. It does nothing to a sealed box on a shelf.
1
12
AI exposure in Europe isn’t where most people expect.
We analyzed 196M EU workers.
Highest exposure:
-Clerks
-IT professionals
-Business administration
Lowest exposure:
Farmers
Cleaners
Manual workers
49M jobs (25%) are highly exposed to AI.
Dataset + methodology
37
30% of restores fail. Most CEOs don’t find out until ransomware hits.
This week HYPERBUNKER was featured in Blocks and Files. The specialist site storage and infra execs read, but one most outside industry have never heard of. Offline. immutable. dumb by design.
1
66
full story
blocksandfiles.com/2025/09/1…
52
£300M. Gone.
Backups weren’t last line. They were entry point.
Attackers phished TCS staff—M&S’s core IT partner.
VMs encrypted via ESXi. No isolation.
TCS lists Veeam, Rubrik, Commvault, and Cohesity as “strategic data protection partners.”
in aviation, postmortems are standard.
1
68
"Chef" scored a better loan than "Head of Culinary Ops."
Same income, same credit history.
The AI favored shorter words.
No punchline. Just how black-box models behave.
1
65
In regulated industries, this isn’t an edge case.
It’s a bomb.
Hariom Tatsat (Barclays) & Ariye Shater call it out in “Beyond the Black Box” (May 2025):
“LLMs in finance require symbolic interpretability to meet audit and compliance needs.”
arXiv:2505.24650
2
33
At @ExtensityAI, we’re building tools that let compliance teams see the reasoning.
Not just outputs—actual symbolic decision chains.
Because explainability isn’t a feature.
It’s survival.
19




