返回新聞
產品展示AI Understanding 簡報

AWS 為 AI 代理程式新增託管 OAuth 同意門戶

Amazon Bedrock AgentCore Identity 現在提供託管同意門戶,將最終用戶 OAuth 授權綁定到個人用戶,並將生成的令牌儲存在令牌庫中。

4 min readRead the primary source
Source-provided image accompanying AWS adds managed OAuth consent portals for AI agents
主要來源文件來源記錄
出版商
aws.amazon.com
來源連結
aws.amazon.comhttps://aws.amazon.com/blogs/machine-learning/manage-end-user-oauth-consent-for-ai-agents-with-amazon-bedrock-agentcore/
來源類型
主要文件-我們直接閱讀的官方公告、文件、文件或第一方頁面。
背景60 秒內了解這一點

從這裡開始

關鍵術語

MCP(模型上下文協定)
一種開放協議,允許人工智慧應用程式以標準方式連接到外部工具、資料來源和上下文提供者。
特點
模型用來進行預測的輸入變數。
代幣
由語言模型處理的文字區塊,例如單字或符號。
測試一下自己AI 代理測驗

發生了什麼事

AWS says AgentCore Identity has added a Consent portal for AgentCore Gateway’s three-legged OAuth flow. Administrators configure an organization identity provider, gateway targets, outbound OAuth providers, and an execution role, then share a portal URL with users. Users sign in, review available services, and authorize providers such as GitHub or Slack independently.

AWS describes the Consent portal as a managed web experience and session-binding endpoint for AgentCore Gateway. The portal handles browser redirects and associates the returning user with the OAuth grant, while AgentCore Identity stores resulting per-user tokens in its vault.

The documented setup requires an administrator to configure a corporate identity provider that issues a JWT access the portal can validate, create or supply an IAM execution role, associate gateway targets with outbound OAuth providers, and configure default return URLs. The source gives examples involving Okta and Auth0 but does not claim that these are the only supported providers.

The end-user flow is provider-specific and independent: a developer opens the shared portal URL, authenticates through the organization’s identity provider, reviews the requested service access, and connects GitHub or Slack. Existing connections remain visible when the portal is reopened unless a grant is revoked, expires, or requires renewed consent. The source highlights use with Kiro, Claude Code, Cursor, and Visual Studio Code, but does not establish that every version or configuration of those clients is supported.

來源詳情: aws.amazon.com ↗

為什麼這很重要

The addresses a practical control problem for AI agents: ensuring that access granted to services such as GitHub and Slack remains tied to the employee who approved it. Previously, customers had to build and host the browser redirects, callback endpoint, user authentication, session handling, and session-binding logic themselves. A managed flow could reduce implementation work and make per-user authorization more consistent for agents used through IDEs and MCP clients, although the source does not establish availability, reliability, or security outcomes beyond AWS’s description.

For an AI coding assistant, per-user OAuth binding determines whose permissions an agent uses when it invokes tools. That is materially different from giving an agent a shared service credential, because the resulting activity can remain associated with the employee who authorized access, subject to the organization’s identity and provider configurations.

The change may lower the amount of custom infrastructure required to deploy agent tool access. AWS specifically says customers previously had to host a public HTTPS callback, manage browser sessions, authenticate the returning user, and call CompleteResourceTokenAuth themselves. The portal centralizes those steps within AgentCore Identity.

CloudTrail visibility provides an operational control: AWS says GetResourceOauth2Token events identify the credential provider, requested scopes, OAuth flow, execution role, and Region, while sensitive and state values are redacted. The source does not independently verify the security of the implementation or quantify any reduction in risk.

Interactive Mechanism

互動機制:它實際上是如何運作的

以互動方式探索這項發展背後的基礎技術。

Agent Lifecycle Stage:
1
User Intent & Planning: "Audit customer refund request #4092 and settle payment."
2
Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').
3
Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.
4
Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.
Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.
互動式概念檢查+10 Points
AI Agents Quiz

An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?

接下來看什麼

Organizations considering the will need to confirm regional availability, pricing, service limits, supported identity providers, and how the managed portal fits their access-review and -lifecycle requirements. AWS’s walkthrough also highlights CloudTrail events for reviewing consent operations and troubleshooting failed token requests.

The source does not provide a price, launch date, service-level commitment, supported AWS Regions beyond saying that portals are listed by account and Region, or a general-availability designation. Access and commercial terms therefore remain unknown.

Administrators must validate the identity provider’s format and carefully scope IAM permissions. The documented setup also creates a dependency between gateway targets and outbound credential providers: AWS instructs administrators to delete the gateway target before deleting its referenced provider.

Organizations should test consent renewal, revocation, expiration, disconnect-and-reconnect behavior, requested scopes, and audit workflows before allowing agents to reach production systems. The source describes these flows but does not report independent testing or user results.

相關指引和測驗

人工智慧代理AI 倫理Prompt Engineering測試你所知道的—嘗試免費的人工智慧測驗在我們的詞彙表中尋找人工智慧術語關注 AI 模型發布追蹤器
覺得有用嗎?