According to The Verge, 1Password has launched a browser integration called 1Password for Claude, which lets users authorize Claude to complete multi-step tasks such as booking travel and managing online accounts without manually entering login credentials (E1). Based on a separate report from iThome, this integration is a joint feature from 1Password and Anthropic that lets Claude log into websites with user authorization, while passwords and one-time codes are never handed to the model, injected into Claude's context, stored in its memory, or sent to Anthropic's systems (E7).
The Verge reports that 1Password built a "zero-exposure security framework" that injects the credentials required for each task through a secure channel the Claude agent cannot view (E2). iThome's account of the same mechanism specifies what stays out of reach: passwords and one-time verification codes are excluded from the model, from Claude's context and memory, and from Anthropic's systems entirely (E7).
Per The Verge, Claude's access is granted per task, and 1Password says users can approve or deny each request with a single biometric prompt (E3). The outlet also quotes 1Password's press release directly: "The moment an AI agent takes control of the browser, 1Password locks down automatically, limiting access to only the credentials explicitly granted for the current task. Nothing else in the 1Password vault is reachable" (E4).
iThome lays out the step-by-step flow: when Claude needs to log into a website, it sends a credential request to 1Password, and the 1Password desktop app displays which login item is being requested and for what purpose. Once the user approves, the browser extension fills the username, password, or one-time code directly into the site — Claude receives only the title of the approved item, the associated username or email, the recorded website, and a success-or-failure status, never the password or the one-time code itself (E8). This matches the mechanism The Verge describes, in which the required credentials are injected through a channel the agent cannot view (E2).
iThome reports that the 1Password browser extension has added an Agentic Mode: when a compatible AI agent takes control of a browser tab, the extension hides autofill suggestions, save-password prompts, and notifications, and the agent can only use the login credentials already approved for the current task — it cannot reach anything else in the vault through the extension (E9). This lines up with 1Password's own statement, quoted by The Verge, that the browser "locks down automatically" the moment an AI agent takes control, restricting it to only the explicitly granted credentials (E4).
The Verge notes that 1Password has not specified exactly which credential types Claude can access, but the access appears limited to login-related details for now, with support for payment cards and identity details promised sometime after launch (E6). iThome's report is more specific about the current scope: 1Password for Claude currently supports usernames, passwords, and one-time verification codes, but does not yet support passkeys (E10).
According to The Verge, 1Password for Claude is available now for 1Password users on Mac, across business, family, and individual plans, and requires both the 1Password desktop app and browser extension as well as the Claude desktop app and browser extension (E5). iThome's report corroborates the platform and plan scope, confirming the feature has launched on Mac for business, family, and individual plans (E10).
iThome reports that 1Password took a different technical approach when it added support for OpenAI Codex in May 2025 (2025-05): through the 1Password Environments MCP Server, secrets are handed directly to the running application at execution time, existing only briefly in memory, and are never written into prompts, code, local files, repositories, or model context (E11). That runtime-injection-into-memory approach differs from the "zero-exposure security framework" 1Password built for Claude, which routes credentials through a secure channel specifically designed to stay outside the agent's view during a live browser task (E2).
| Integration | Partner | Timing | Mechanism | What the AI never sees |
|---|---|---|---|---|
| 1Password for Claude | Anthropic | Available now (Mac; business/family/individual plans) | Zero-exposure security framework injects credentials via a secure channel during browser tasks | Passwords, one-time codes (E2, E5, E7) |
| OpenAI Codex support | OpenAI | May 2025 | 1Password Environments MCP Server injects secrets into the running application at execution time | Secrets in prompts, code, local files, repositories, model context (E11) |
Across both The Verge's and iThome's reporting, the common thread is that 1Password's design goal for Claude is the same one it already tried with OpenAI Codex in May 2025: keep secrets out of the model's reach entirely, whether that's a password field during a live browser session or an API key during code execution (E11, E2). The specific mechanics differ — one is a secure-channel injection during a browser task with per-task biometric approval (E2, E3), the other is a memory-only handoff to a running application (E11) — but in both cases 1Password positions itself, not Anthropic or OpenAI, as the party that ever touches the actual credential (E7). The current gap in the iThome and Verge accounts is scope: passkeys, payment cards, and identity details are explicitly not yet supported for Claude (E10, E6), and the feature is confirmed only on Mac so far (E5, E10) — so the vault-lockdown design described in both reports currently covers a narrower set of credential types and platforms than 1Password's password-and-OTP-based login flow.
Not yet. iThome reports the feature currently supports usernames, passwords, and one-time verification codes, but does not yet support passkeys (E10).
According to iThome, Claude only receives the approved item's title, the associated username or email, the recorded website, and a success-or-failure status — never the password or one-time code itself (E8).
The Verge reports it is available now on Mac, across business, family, and individual 1Password plans, requiring both the 1Password and Claude desktop apps and browser extensions (E5).
According to reports by CNA, CTEE and UDN, TSMC (台積電) will add $100 billion to its Arizona investment, lifting the total to $265 billion, while raising 2026 capex to $60–64 billion. CFO Wendell Huang (黃仁昭) called AI chip demand a multi-year structural trend, though a TSMC executive flagged local construction-worker shortages as a challenge.
According to Cnyes and MoneyDJ, Kung Sheng Insurance Broker (公勝保經) topped brokerage evaluations for 7–8 consecutive years, posted record 2025 revenue of NT$4.715 billion and EPS of NT$7.86, reached 631 MDRT members, and listed on the Taipei Exchange (TPEx) on March 30, 2026.
According to TechNews, Hugging Face confirmed its production environment was breached by an AI agent-led attack that stole internal datasets and credentials via two abused code-execution paths, leaving over 17,000 event log entries. iThome reports the agent escalated to cluster-level access and moved laterally into multiple internal clusters within a single weekend. Hugging Face says it found no evidence of tampering with models, datasets, or its software supply chain.