Claude Duet

Anthropic describes Duet’s dedicated hosted computer/browser, which establishes the product’s cloud operation. The reviewed docs do not quantify the service’s energy, identify electricity sourcing for the compute facility or offer workload resource controls tied to physical reductions; no impact is inferred.

1 + 1 reporting + 0 electricity + 0 controls = 2/10. This grades documented practices and transparency.

Task energy, serving model route, region/facility electricity and compute resource controls are not reported for Duet.

Reporting1/3

Anthropic describes a separate cloud computer and browser for Duet, establishing hosted operation but no task energy reporting.

Electricity sourcing0/3

The Duet cloud-computer description does not map execution to a named facility or disclose that facility’s electricity mix.

Resource controls0/3

The reviewed Duet sources provide no user-visible token, context or compute control documented as reducing physical resource demand.

• Introducing Claude Duet ↗
• Claude Duet Safety, Privacy, and Security ↗
• Get started with Claude Duet ↗

Anthropic’s Duet help page identifies Opus 5.5, whose Transparency Hub describes a proprietary mix of publicly available online information, public/private datasets, user data and synthetic data. It provides no named-source rights ledger; account training preferences are distinct from the training data used to train the model.

1 + 1 source transparency + 0 permission + 0 creator controls = 2/10, with the permission cap applied.

No complete source/license inventory or Opus 5.5-specific evidence of creator removal and retraining was found in the reviewed disclosures.

Source transparency1/3

Anthropic’s Opus 5.5 transparency entry describes a proprietary mixture of broad data categories without a named-source inventory; Duet’s help page names Opus 5.5.

Documented permission0/4

The disclosed broad data categories do not document an affirmative rights basis for the material training data.

Creator controls0/2

The reviewed model transparency material doesn’t show a training-data creator opt-out or removal-and-retraining process for Opus 5.5.

• Get started with Claude Duet ↗
• Anthropic Transparency Hub ↗
• Anthropic use of personal data in model training ↗

Anthropic says Duet inherits Claude memory, skills and connected apps, stores task history and memory until deletion/reset, and runs its own cloud computer and browser. Its design keeps passwords and tokens outside the agent machine and adds an external approval/classifier layer, but this remains hosted processing.

5/10 reflects documented task approvals, isolated credentials and reset/disconnect controls, while the research preview remains cloud-hosted and inherits account-connected data.

Retention periods, region-specific processing, model-training application and account controls follow Claude plan terms and the research-preview settings.

• Get started with Claude Duet ↗
• Claude Duet Safety, Privacy, and Security ↗

Anthropic identifies Claude Opus 5.5 as the Duet model and describes a private cloud agent/browser architecture. The cited materials do not make the runtime or model weights available for inspection or self-hosting, although the product has connectors and user controls.

2/10 reflects a managed, waitlisted cloud preview with a fixed model and no published source license or user-controlled deployment route.

The service implementation and full model training record are not exposed in the cited launch materials.

• Introducing Claude Duet ↗
• Get started with Claude Duet ↗

Anthropic’s July 2026 Transparency Hub says suppliers must comply with a Vendor Code of Conduct covering labor and human rights and apply equivalent standards through their own supply chains; the Trust Center lists a Global Vendor Code of Conduct, but its text is access-gated in the public interface. This supports limited credit for a supplier standard, not a finding about annotator pay, exposure controls or actual enforcement. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 1 pay & conditions + 0 worker voice & remedy + 0 evidence & follow-through = 2/10. This grades documented worker safeguards, not a measured rate of mistreatment.

The public code text, worker complaint outcomes, rates/pay, annotation supplier identities and workforce coverage were not accessible in this review; company model documentation doesn’t confirm current worker treatment.

Pay & conditions1/3

Anthropic states that its supplier code sets labor and human-rights requirements and cascades equivalent standards through vendor supply chains; the publicly listed code text is access-gated, so no specific working conditions or application are verified.

Worker voice & remedy0/3

No public worker-specific grievance channel, opt-out, worker survey or worker representation mechanism was established in the sources reviewed.

Evidence & follow-through0/3

The Trust Center lists a vendor code and security audits for vendors handling Anthropic data, but no accessible evidence of labor audits, worker remediation or outcomes was found.

• Anthropic Transparency Hub: Voluntary Commitments ↗
• Anthropic Trust Center: Global Vendor Code of Conduct ↗

Anthropic’s Feb. 26, 2026 statement says Claude was extensively deployed in classified Department of War and national-security settings for intelligence analysis, modeling and simulation, operational planning and cyber operations. The same statement says Anthropic had not objected to particular operations ad hoc and that lawful foreign-intelligence use was supported, while opposing mass domestic surveillance and fully autonomous weapons. This confirms broad defense enablement and named military workflow use, but does not identify target selection, strike decisions or combat-intelligence support. The statement reports the DoW was threatening offboarding; whether the deployment/contract continued by the October 2026 review date is not publicly established in the sources reviewed. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 1 use restrictions + 1 enforcement & oversight + 2 relationship disclosure = 5, capped at 3/10. Known defense ties lower this rating; safeguards can improve it within the relationship’s cap.

The public statement does not enumerate active task orders, system versions, unit-level deployments, operational outcomes, or independent audit results. DoW supply-chain risk litigation and court orders make access status time-sensitive; the company's current page describes the status as of its stated dates.

Use restrictions1/4

Anthropic says it opposed mass domestic surveillance and fully autonomous weapons in its DoW contract discussions, but supports lawful foreign intelligence and partially autonomous weapons; the statement reports the DoW sought broader any-lawful-use terms.

Enforcement & oversight1/3

Anthropic states that it has safeguards and resists certain uses, but the reviewed statement provides no independent military-use review, published enforcement or remedy outcomes.

Relationship disclosure2/2

Anthropic names classified deployments and use categories and reports a dispute and possible offboarding; it does not disclose actual missions, current use-level limits, or post-dispute contract status.

Anthropic reported mission-critical classified use across multiple military workflows, but the source doesn’t confirm target selection, strikes, or combat intelligence; broad-defense cap applies. Current post-dispute contract status is unknown. The relationship limits this grade to 3/10.

• Statement from Dario Amodei on our discussions with the Department of War ↗

The official BNC guide and No Tech campaign list named corporate targets; Anthropic does not appear in the reviewed target lists. This limited check is not an endorsement and doesn’t show relationships with cloud or government customers.

This is limited to the two official BNC campaign sources; it does not cover every local campaign, supplier, investment, or future change.

• BNC consumer boycott and pressure guide ↗
• BNC No Tech campaign ↗