LinkedIn·Friday, 14 August 2026·12d ago
July has been interesting for #CyberSec & #AI, frontier AI labs disclosed that models under evaluation escaped their sandboxes and reached…
Nathan Bijnens
CEO at Scrydon | xMSFT | Your Data, Your AI, Your Control
July has been interesting for #CyberSec & #AI, frontier AI labs disclosed that models under evaluation escaped their sandboxes and reached real third-party infrastructure.
As Scrydon our starting assumption always is Treat the Model as an Untrusted Workload. Great write-up on how we tackle this by Xavier Geerinck!
♥ 12↻ 1
View on LinkedIn Cross-referenced
Related on the wire
Every time my house decided to charge its battery to dodge Flanders’ capacity tariff, that command travelled from Belgium to a…
How a version update can cause major impact... in a good way!
Five people decide whether AI actually happens in a European public body. 𝗧𝗵𝗲𝘆 𝗮𝗹𝗺𝗼𝘀𝘁 𝗻𝗲𝘃𝗲𝗿 𝘀𝗶𝘁 𝗶𝗻 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲…
𝗪𝗵𝗮𝘁 𝗣𝗮𝗹𝗮𝗻𝘁𝗶𝗿 𝗚𝗼𝘁 𝗥𝗶𝗴𝗵𝘁 𝗪𝗶𝘁𝗵 𝗢𝗻𝘁𝗼𝗹𝗼𝗴𝗶𝗲𝘀 They put the #ontology at the centre and the model at the edge —…
Governance as kernel, Context as memory, Process & Workflows as scheduler, Integrations as drivers — and Humans, deliberately last, as the…
𝗪𝗵𝗼 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗻𝗲𝗲𝗱𝘀 𝗮𝗻 𝗔𝗜 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗦𝘆𝘀𝘁𝗲𝗺? Most companies are still treating AI like a fancy personal…