EU AI Act Omnibus Delays High-Risk Deadline; AutoGen Enters Maintenance Mode

<p>The EU Digital Omnibus pushed the high-risk AI compliance deadline to <strong>December 2, 2027</strong> — but GPAI fine enforcement still activates August 2, so model providers need documentation ready now. Meanwhile, Microsoft moved AutoGen to maintenance-only status and redirected teams to the <strong>Microsoft Agent Framework</strong>, while CrewAI ...

Highlights

Today's leading signals

Lead signal · Governance

The EU Digital Omnibus pushed the high-risk AI (Annex III) compliance deadline from August 2, 2026 to December 2, 2027 — but GPAI fine enforcement still activates August 2, making model-provider obligations the live near-term priority. (EU Council)

Formal Parliamentary adoption must still clear before August 2; compliance program structures should stay in place. (Gibson Dunn)
Microsoft placed AutoGen in maintenance mode, redirecting teams to the Microsoft Agent Framework — a merge of AutoGen and Semantic Kernel with an LTS commitment. (Microsoft Learn)
CrewAI ships native MCP and A2A protocol support; LangGraph teams still rely on community integrations for the same interoperability. (CrewAI)
Key Signals

What changed

  1. EU AI Act Omnibus: High-Risk Deadline Moves to 2027; GPAI Enforcement Does Not May–June 2026

    The Council and Parliament reached a provisional Omnibus agreement on May 7, 2026, moving Annex III stand-alone high-risk system obligations to December 2, 2027 (Annex I product-embedded systems shift to August 2, 2028). But August 2, 2026 still carries real weight: that is when the EU AI Office's power to impose fines on GPAI model providers activates. Teams that are providers of general-purpose AI models — not merely deployers — need technical documentation and model factsheets ready before that date, regardless of the Omnibus delay. (EU Council, Gibson Dunn)

  2. AutoGen in Maintenance Mode; Microsoft Agent Framework Is the Successor May 2026

    Microsoft's May 2026 memo formally put AutoGen on a maintenance-only footing — critical security patches only, no new features, community-managed going forward. The replacement, the Microsoft Agent Framework, merges AutoGen's agent abstractions with Semantic Kernel's enterprise capabilities: session-based state, typed middleware, telemetry hooks, and graph-based multi-agent workflows, with MCP and A2A support included. Teams with AutoGen in production need a migration timeline before community forks diverge. (Microsoft Learn)

  3. CrewAI's Native MCP and A2A Support Widens the Framework Selection Gap Spring 2026

    CrewAI now ships first-class MCP tool-server connectivity and A2A (Agent-to-Agent Protocol) support natively, while most orchestration frameworks still rely on community wrappers for the same capability. For mixed-runtime pipelines, native A2A reduces the hand-rolled bridging that has been one of multi-agent production's persistent pain points. LangGraph's strengths remain LangSmith observability and state persistence — the protocol-native gap is now a real selection criterion. (CrewAI)

What To Watch

Why It Matters

  1. August 2 still has teeth — know which side of the provider/deployer line you stand on
    • Model providers above the 10²⁵ FLOP systemic-risk threshold face live EU AI Office fine enforcement in 38 days — finalize technical documentation and model factsheets now. (Gibson Dunn)
    • High-risk system deployers (HR screening, credit scoring, education platforms) gain until December 2027 — but only after formal Omnibus adoption clears. (EU Council)
  2. AutoGen users need a migration decision before the ecosystem forks
    • Audit graph nodes for abstractions that won't receive compatibility shims; Microsoft's migration guide maps the full API surface. (Microsoft Learn)
    • CrewAI's protocol-native support and LangGraph's LangSmith integration are the strongest active-development alternatives. (CrewAI)
  3. Protocol stance is becoming a first-class framework selection criterion
    • Native MCP support reduces brittle adapters between agents and tool ecosystems; evaluate your framework's MCP and A2A story before committing to a multi-agent architecture. (CrewAI)
Sources

Quick Links