Anthropic details Claude Duet's approval cards, sensitive-info classifier and credential isolation
Anthropic describes safeguards for Claude Duet, an agent in research preview that runs on a cloud computer and acts for users across connected apps. Before sending sensitive messages, paying, or deleting, it shows users the action and waits for approval, with a separate automated check outside the model. A classifier screens outgoing messages for sensitive information, sorted into never-share and needs-consent tiers. Credentials are kept away from the model: passwords are encrypted for single use, and sign-in tokens are swapped in by an outside proxy. The article also lists limitations, noting the safeguards will sometimes miss things or flag harmless messages.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · automated brief
- Discovery linkX
- Linked in sourceClaude Duet (opens in new tab)
- Linked in sourcepublished our framework for safe and trustworthy agents (opens in new tab)
- Linked in sourceno single defense is enough (opens in new tab)
- Linked in sourceprotection inside the model can’t stand alone (opens in new tab)
- Linked in sourceClaude Managed Agents (opens in new tab)
- Linked in sourcethe auto mode classifier from Claude Code (opens in new tab)
- Linked in sourcebrowser-use research (opens in new tab)
- Linked in sourceHackerOne (opens in new tab)
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
Anthropic describes safeguards for Claude Duet, an agent in research preview that runs on a cloud computer and acts for users across connected apps. Before sending sensitive messages, paying, or deleting, it shows users the action and waits for approval, with a separate automated check outside the model. A classifier screens outgoing messages for sensitive information, sorted into never-share and needs-consent tiers. Credentials are kept away from the model: passwords are encrypted for single use, and sign-in tokens are swapped in by an outside proxy. The article also lists limitations, noting the safeguards will sometimes miss things or flag harmless messages.
Key findings from the stored brief
- Claude Duet shows users the action, destination and content, including all recipients, and waits for approval before sensitive steps.
- A separate classifier checks outgoing messages and blocks never-share items such as credentials, card and bank numbers, and government IDs.
- Passwords are encrypted with a single-use key and passed to the browser unopened, so the model never sees them.
- Anthropic states the safeguards will occasionally miss something or flag a harmless message, and the classifier is still being iterated.
Source claims not independently checked; the article describes a research preview whose safeguards Anthropic says are still being calibrated.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
Agents that act for users while they are away raise privacy and security risks. The article describes concrete design choices, including approval gates, outside-model checks and credential isolation, that other agent builders could evaluate.
Limits and open questions
Source claims not independently checked; the article describes a research preview whose safeguards Anthropic says are still being calibrated.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.