EU AI Act: 40 Days to Comply; Claude Code Redesigns Agent Teams

<p>The <strong>EU AI Act's high-risk deadline hits August 2</strong> — 40 days out — requiring agent builders to register, document decision logic, and wire in human oversight before penalties of up to €15M kick in. Meanwhile, <strong>Claude Code overhauled multi-agent teams</strong> with implicit team sessions and a new <code>Tool(param:value)</code> per...

Today’s highlights

Highlights

EU AI Act high-risk deadline: 40 days out

The EU AI Act’s high-risk AI system deadline lands August 2 — requiring registration, risk documentation, and structured human oversight for any agent making consequential autonomous decisions in EU contexts. (IAPP)

governance eu-ai-act 40-day window compliance

Claude Code multi-agent overhaul

TeamCreate/TeamDelete replaced by implicit teams; new Tool(param:value) permission syntax ships. (Anthropic)

Vertex AI SDK cutover live

vertexai.* imports throw ImportError today. Migrate to google-genai now. (Google Cloud)

Fable 5 & Mythos 5 still suspended

12 days offline under US export-control directive; no restoration timeline. Confirm alternative model routing before the August audit window. (Fortune)

Key signals

Key Signals

  1. 1 EU AI Act high-risk deadline: 40 days out August 2, 2026

    Full high-risk AI system obligations take effect August 2, including mandatory EU database registration, technical documentation of decision logic, open-loop architecture with human-override points, and evidence of ongoing risk management. Agents that determine credit scores, screen job applicants, handle regulatory reporting, or make autonomous infrastructure decisions are squarely in scope — penalties reach up to €15M or 3% of global revenue for violations. Many SMEs have not completed deployer evidence requirements and are running out of runway. (IAPP)

  2. 2 Claude Code agent teams redesign ships June 2026

    Anthropic removed the explicit TeamCreate/TeamDelete lifecycle tools. With CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, every session now carries one implicit team and teammates are spawned directly via the Agent tool’s name parameter — no setup step. Nested skills in .claude/skills/ load automatically when working in that directory tree; name clashes resolve as <dir>:<name>. The new Tool(param:value) rule syntax — e.g., Agent(model:opus) to block Opus subagents — gives platform operators a model-governance lever inside multi-agent pipelines. (Anthropic)

  3. 3 Vertex AI SDK is now gone June 24, 2026

    Google’s one-year deprecation window expired today. vertexai.generative_models, vertexai.language_models, vertexai.vision_models, vertexai.tuning, and vertexai.caching are removed from google-cloud-aiplatform; affected production code throws ImportError at startup. The replacement is google-genai (google.genai.Client), which unifies the Gemini API and Vertex AI backends. The Vertex AI console disappeared May 21; today completes the platform’s full transition to Gemini Enterprise Agent Platform. (Google Cloud)

Operator actions

Why It Matters / What To Watch

  1. EU AI Act compliance: start the audit this week
    • Forty days is workable if you begin immediately — audit agents for autonomous decision scope (employment, credit, infrastructure) before anything else. (IAPP)
    • Shadow AI deployments — teams running agents without centralized oversight — are non-compliant under the Act; platform operators should inventory all agent deployments by department before August 2. (Secure Privacy)
  2. Claude Code permission rules unlock model-level budget governance
    • The Tool(param:value) syntax is directly useful for operators controlling cost in multi-agent pipelines: Agent(model:opus) blocks expensive subagent calls without touching application logic. (Anthropic)
    • Any team using MCP servers with authentication in headless or SDK mode should retest after the update — the auth-stub exposure bug affected tool visibility in those deployment shapes. (Anthropic)
Sources

Quick Links