https://packetsherpa.org/tags/ai/Recent content in Ai on Default Deny, Thoughts from a Packetsherpahttps://packetsherpa.org/images/og-default.jpghttps://packetsherpa.org/images/og-default.jpgHugoen-USSat, 01 Aug 2026 12:24:15 -0400https://packetsherpa.org/anthropic-missed-the-security-lesson/Sat, 01 Aug 2026 12:24:15 -0400https://packetsherpa.org/anthropic-missed-the-security-lesson/One lab's model used a zero-day, another lab discovered three or their models used an open path and is litigating that distinction when the common denominator is the one that matters: nothing in the path applied policy to outbound traffic.https://packetsherpa.org/shadow-ai-is-a-demand-signal-not-a-policy-failure/Thu, 30 Jul 2026 22:29:11 -0400https://packetsherpa.org/shadow-ai-is-a-demand-signal-not-a-policy-failure/Blocking an unsanctioned AI tool doesn't remove the need behind it. It only removes your visibility into how employees are already solving it.https://packetsherpa.org/it-wasnt-air-gapped/Fri, 24 Jul 2026 00:00:00 -0400https://packetsherpa.org/it-wasnt-air-gapped/A model can exploit a path it can reach. It cannot exploit a route that does not exist.