https://yesiddo.com/en/topics/ai/Cybersecurity, AI and coding with agents: what I read, test and keep at hand.en-USWed, 14 Oct 2026 11:00:00 +0200https://yesiddo.com/en/notes/ai-pulse-2026-sovereignty-and-open-models/Wed, 14 Oct 2026 11:00:00 +0200notes/ai-pulse-2026-souverainete-et-modeles-ouverts/index.en.mdThe word sovereignty was everywhere at Station F, right down to the signage. Behind the slogan, a fairly coherent strategy: open weights, GPUs in Europe and the middle layer.https://yesiddo.com/en/notes/ai-pulse-2026-agent-identity/Wed, 14 Oct 2026 10:00:00 +0200notes/ai-pulse-2026-identite-des-agents/index.en.mdZalando runs agents across 2,700 microservices. Their answer to scattered secrets is open source, and it fits in one RFC.https://yesiddo.com/en/notes/ai-pulse-2026-inference-economics/Wed, 14 Oct 2026 09:00:00 +0200notes/ai-pulse-2026-economie-de-l-inference/index.en.mdAt ai-PULSE 2026, AMD talked about reselling tokens and Nokia about networking inside the racks. Read together, the two talks show that what costs money is the time the chip is not working.https://yesiddo.com/en/notes/ai-pulse-2026-takeaways/Wed, 14 Oct 2026 08:00:00 +0200notes/ai-pulse-2026-compte-rendu/index.en.mdA report on Scaleway's AI conference at Station F: the opening keynote, sovereignty, inference economics, agent identity. With links to the detailed write-ups.https://cyber.gouv.fr/sites/default/files/document/Recommandations_de_s%C3%A9curit%C3%A9_pour_un_syst%C3%A8me_d_IA_g%C3%A9n%C3%A9rative.pdfMon, 12 Oct 2026 10:00:00 +0200liens/anssi-securite-ia-generative.en.md<p>35 recommendations, from training to production: manipulation, infection, exfiltration. The linked document is in French.</p>https://www.anthropic.com/engineering/building-effective-agentsMon, 12 Oct 2026 09:40:00 +0200liens/building-effective-agents.en.md<p>Workflow or agent? Start simple, and add autonomy only when it really pays off.</p>https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/Mon, 12 Oct 2026 09:10:00 +0200liens/lethal-trifecta.en.md<p>Private data, untrusted content, an exfiltration channel: the combination an agent must never bring together.</p>https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/Mon, 12 Oct 2026 09:00:00 +0200liens/owasp-top-10-agentic-2026.en.md<p>Goal hijacking, tool misuse, memory poisoning: the list of risks specific to autonomous agents.</p>https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agentsMon, 12 Oct 2026 08:50:00 +0200liens/context-engineering.en.md<p>Context is a limited resource: what you give an agent to read matters as much as what you ask of it.</p>https://modelcontextprotocol.io/Mon, 12 Oct 2026 08:40:00 +0200liens/model-context-protocol.en.md<p>The standard protocol for plugging tools and data into an agent. Read it with the &ldquo;lethal trifecta&rdquo; in mind.</p>