🧭 Claude Academy Is Live — Free AI Education for Everyone, No Sign-In Required

Anthropic launched Claude Academy on August 20, 2026 — a free, publicly accessible learning platform at anthropic.com/learn that covers how to use Claude safely and effectively, AI fluency fundamentals, and (for developers) how to build agents with the Claude API. No Claude account or sign-in is required to access the course content, though registered users can track their progress and earn LinkedIn-shareable completion badges.

What Claude Academy contains

• AI Fluency: Framework & Foundations — a 4-hour, 14-lesson course built around Anthropic's 4D AI Fluency Framework: Delegation (which tasks to hand off), Description (how to articulate tasks clearly), Discernment (evaluating AI output critically), and Diligence (verifying and taking responsibility for the result). This is the most immediately useful track for business users who are just starting.
• Claude Essentials — a shorter track covering the Claude interface, Projects, Memory, and Cowork, aimed at everyday users on any plan.
• Build with Claude — developer-focused content on the Claude API, tool use, MCP integration, Claude Code, and multi-agent patterns. This section mirrors the structure of the developer platform docs but adds interactive exercises and worked examples.
• Responsible AI use — content on recognising AI limitations, avoiding over-reliance, and using Claude within appropriate governance frameworks.

Why this matters beyond the obvious

Claude Academy effectively consolidates three previously fragmented things: the old Anthropic Academy (13 developer courses on a third-party LMS, launched quietly in March 2026), the support documentation, and the product marketing "use case" pages. Pulling these into a unified, credentialed platform means organisations can now point new staff at a single URL for onboarding — and managers can verify completion. The LinkedIn badge integration is a small but meaningful signal that Anthropic is treating AI fluency as a professional credential, not just product training.

🧭 Claude for Teachers Expands to Full School Districts — Free Enterprise for US K-12

Anthropic made Claude for Teachers available to US K-12 schools and districts as a free Enterprise offering on August 28, 2026, extending a programme it originally launched for individual verified educators on July 14. The district-level tier brings Claude's full Enterprise feature set — SSO, role-based access controls, shared usage policies, and an admin console — to school systems at no cost, with one year of free access guaranteed for qualifying organisations that sign up before June 30, 2027.

What schools and districts get

• Full Enterprise access for all staff: Administrators can bring every teacher and support staff member into a single organisation under one policy set. All seats are Enterprise-grade — the same tier that large corporations pay for — not a stripped-down "education plan."
• SSO and role-based controls: Integration with the school's existing identity provider (Google Workspace for Education, Microsoft Entra, Clever) means staff do not need separate Anthropic credentials, and admins can enforce which features are available to which roles.
• Teaching-specific skills grounded in learning science: The Claude for Teachers edition ships with skills that are aware of US academic standards across all 50 states, can draft differentiated lesson plans and IEP accommodation notes, and are calibrated for age-appropriate language when scaffolding student assignments — distinct from the generic Claude model behaviour.
• Detroit Public Schools pilot: Anthropic is running a formal evaluation of the programme with the Detroit Public Schools Community District this fall, studying impacts on educator wellbeing and classroom practice. Results are expected to inform future product direction.

🧭 Managed Agents Get Hard Spend Caps and Geo-Pinned Inference

Two developer platform updates have shipped for Claude Managed Agents that address two of the most common operational concerns teams raise when moving agents from staging to production: runaway spend and data residency. The changes are live in the Claude developer platform and configurable via the Managed Agents API.

Budget controls: hard cap on session spend

You can now set a budget on a Managed Agents session — a hard ceiling on how much that session can spend, priced at public list rates. When the session hits the cap, it pauses cleanly with a budget_reached stop reason rather than erroring out. You can then raise or remove the budget to resume the session without losing its state. Deployments also accept the same budget parameter and apply it as a per-session default for every session they start.

# Python — set a $1.50 budget on a Managed Agents session
session = client.beta.managed_agents.sessions.create(
    agent_id="agt_xxx",
    budget={"amount": 1.50, "currency": "usd"}
)

# If paused:
# session.stop_reason == "budget_reached"

# Resume by raising the budget:
client.beta.managed_agents.sessions.update(
    session.id,
    budget={"amount": 5.00, "currency": "usd"}
)

The budget is evaluated at model-request boundaries, not mid-generation, so a session will not be cut off in the middle of a streaming response. This matters for agents running long tool-chain sequences: the pause happens after the current turn completes, giving you clean state to inspect before deciding whether to continue.

Geo-pinned inference: keep model execution in-region

The inference_geo parameter on the agent's model object lets you control where model inference runs. Setting it to "us" keeps all model execution within US data centres; "global" (the default) routes to wherever there is capacity. In-region inference is billed at 1.1× the standard rate — a 10% premium for the regulatory guarantee.

# Create an agent pinned to US inference
agent = client.beta.managed_agents.agents.create(
    model={
        "name": "claude-fable-5-1",
        "inference_geo": "us"
    },
    ...
)

# Or override on a per-session basis:
session = client.beta.managed_agents.sessions.create(
    agent_id=agent.id,
    model={"inference_geo": "us"}
)