LIVE
Anthropic says Claude helped trace the origin of NIST's mysterious ECDSA seeds16/10/26 · Anthropic|Reflection AI unveils Beam, a 501-billion-parameter open model16/10/26 · Reflection AI|Meta releases Muse Spark Open, a mixture-of-experts model built for local Mac Studio inference15/10/26 · Meta|Meta releases Muse Spark Open, an open-weight multimodal model15/10/26 · Meta|Google DeepMind outlines safety measures behind Gemini 4 Argon13/10/26 · Google DeepMind|Tev-Gym: A Training Recipe to Fix Sparse Rewards in Long-Horizon Agents13/10/26|Google DeepMind references Gemini 4, code-named Argon13/10/26 · Google DeepMind|Google DeepMind unveils Gemini 4 Argon, a robotics-focused variant13/10/26 · Google DeepMind|Character.AI files for Chapter 11 bankruptcy, sells assets to Disney13/10/26 · Character.AI|Study Shows AI Models Can Provably Conceal Arbitrary Capabilities13/10/26|NVIDIA RTX Spark laptops bring local AI to premium Windows PCs17/10/26 · NVIDIA|OpenAI discloses four more incidents involving sensitive data17/10/26 · OpenAI|Anthropic says Claude helped trace the origin of NIST's mysterious ECDSA seeds16/10/26 · Anthropic|Reflection AI unveils Beam, a 501-billion-parameter open model16/10/26 · Reflection AI|Meta releases Muse Spark Open, a mixture-of-experts model built for local Mac Studio inference15/10/26 · Meta|Meta releases Muse Spark Open, an open-weight multimodal model15/10/26 · Meta|Google DeepMind outlines safety measures behind Gemini 4 Argon13/10/26 · Google DeepMind|Tev-Gym: A Training Recipe to Fix Sparse Rewards in Long-Horizon Agents13/10/26|Google DeepMind references Gemini 4, code-named Argon13/10/26 · Google DeepMind|Google DeepMind unveils Gemini 4 Argon, a robotics-focused variant13/10/26 · Google DeepMind|Character.AI files for Chapter 11 bankruptcy, sells assets to Disney13/10/26 · Character.AI|Study Shows AI Models Can Provably Conceal Arbitrary Capabilities13/10/26|NVIDIA RTX Spark laptops bring local AI to premium Windows PCs17/10/26 · NVIDIA|OpenAI discloses four more incidents involving sensitive data17/10/26 · OpenAI|
+ nAIvigateA SHEET FOR PEOPLE WITHOUT 2 HOURS
+ nAIvigate · fiches / ai-act-aout-2026
▣ fiche-001.md ×
□ index.md
☰ ContentsVerdictTimelineEnforceable nowAction Plan

AI Act: postponed? Not the way you think.

The Digital Omnibus (EU) 2026/1744 has pushed high-risk obligations back to 2027–2028. But since 2 August 2026, authorities have held their sanctioning powers, and transparency is mandatory. “It's been postponed” is the most expensive phrase in your boardroom.

Verdict

the essentials in 10 seconds3 rows
ACTIVEArticle 50 transparency, sanctioning powers, regulatory sandboxes.NOW
POSTPONEDAnnex III high risk → Dec 2027 · Annex I → Aug 2028. Postponement ≠ exemption.OMNIBUS
NEW2 practices added to the blacklist, maximum sanction tier.DEC 2026

Timeline

what's due, and when6 rows
02.02.2025Prohibited practices + AI literacy. Already enforceable, often overlooked.IN FORCE
02.08.2025GPAI models, unchanged, unaffected by the omnibus.IN FORCE
02.08.2026Sanctions + Article 50 transparency. Authorities operational.YOU ARE HERE
02.12.2026New prohibitions + end of the marking window for existing AI content stock.D−107
02.12.2027Annex III high risk: HR, scoring, education, critical infrastructure.MOVED
02.08.2028Annex I high risk: AI embedded in regulated products.MOVED

Enforceable now

the real scope of risk5 rows
art-50User information: anyone facing an AI must be told so.ENFORCEABLE
marquageIdentifiable generated content: published texts, images, videos.ENFORCEABLE
contrôleAuthorities' powers: investigations, orders, fines.ENFORCEABLE
littératieStaff training: the number-one blind spot for SMEs and mid-caps for the past 18 months.ENFORCEABLE
gpaiModel regime: in force since August 2025.ENFORCEABLE
€35 MOR 7% OF GLOBAL TURNOVER

Maximum tier, active since 2 August. Postponing high risk has disarmed neither the authorities nor the fines on what remains enforceable.

Action Plan

to launch this week5 rows
01Inventory actual AI usage: including embedded SaaS and Shadow AI.WITHIN 10 DAYS
02Clear notice on every chatbot: internal as well as external.
03Mark published content, new: immediate, existing stock: window until Dec 2026.DEC 2026
04Document the training: a written record is worth more than an intention.
05Plan for high risk: Dec 2027 is coming fast. No panic, no shelving it.

Go deeper

fundamentals, in Learn2 links
→Agents IA autonomes : le bulletin ANSSI décrypté18 MIN→Architecture agentique : comprendre avant de déployer25 MIN
Don't know what's running in your organisation? Shadow AI Sentinel: 90 days to map your AI usage, declared and undeclared.
REQUEST AN AI RADAR →
⎇ mainfiche #001 · reglementairesources: JOUE 24.07.2026 · (UE) 2026/1744 · (UE) 2024/1689
Follow nAIvigate | naivigate.com/en/fiches