Whisper speech recognition

This downloadable model profile scores disclosed deployment choices and operating controls, not measured energy savings. Local inference lets the operator select device, runtime and workload size, although the model card does not report task-level electricity.

1 + 1 reporting + 0 electricity + 2 controls = 4/10. This grades documented practices and transparency.

Comparable per-minute or per-generation electricity, water use, lifecycle emissions and independently verified clean-power sourcing are not published in these sources.

Reporting1/3

The model and local route are documented, but no environmental footprint is audited.

Electricity sourcing0/3

No attributable clean-electricity procurement or hourly matching evidence was found in the cited primary material.

Resource controls2/3

Choosing a small model and local runtime allows workload and hardware controls; it is not measured energy reduction.

• OpenAI Whisper official repository ↗
• OpenAI official company, product or policy source ↗

OpenAI’s Whisper model card says the release trained on 680,000 hours of audio and transcripts collected from the internet, with broad language proportions. It does not list source works, licenses or speaker consent. The MIT license applies to the code and weights, not to the original training recordings.

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

The full source inventory, speaker consent records, exact licenses for the remaining training data, rights-holder removal process and independent rights audit are not published in the reviewed model materials.

Source transparency1/3

The card discloses the training data scale and broad web-derived source class, but no itemized source inventory.

Documented permission0/4

Internet collection is described without affirmative source-level permission, license or consent evidence.

Creator controls0/2

No rights-holder opt-out, removal or retraining remedy is documented in the model release sources.

• OpenAI Whisper model card ↗

The published model can be downloaded and run locally. Once downloaded, inference need not send recordings or transcripts to the model publisher; software dependencies, telemetry, backups and any remote application remain separate data paths under the operator’s control.

The model publisher does not audit the user’s runtime, network connections, logging, backups or downstream application. Verify those components for your installation.

• OpenAI Whisper official repository ↗
• OpenAI official company, product or policy source ↗

OpenAI publishes Whisper code and model weights under MIT and provides inference instructions. This supports local deployment, while training data disclosure remains broad rather than a fully reproducible manifest.

A complete data manifest, exact training recipe and independently reproduced end-to-end training are not supplied.

• OpenAI Whisper official repository ↗

The October 2026 Supplier Code applies to suppliers, staffing agencies, subcontractors and sub-tier suppliers, and explicitly covers annotation, moderation, evaluation and RLHF workers: it requires task-risk notice, breaks, exposure limits/support, task opt-out or escalation without retaliation, reasonable paid performance expectations, and supplier records that OpenAI may validate. A 2023 TIME investigation into the earlier Sama contract documented workers’ accounts of very low take-home pay and distressing work; Sama disputed key characterizations and said it offered extensive counseling, while OpenAI said it had understood workers could opt out and access support, and said vendors set pay and care. These historical claims and counterclaims do not verify the current code’s implementation, worker voice or outcomes. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 1 pay & conditions + 1 worker voice & remedy + 1 evidence & follow-through = 4/10. This grades documented worker safeguards, not a measured rate of mistreatment.

The current code does not publish supplier-level audit results, pay distributions, annotator coverage, worker survey results or independent remediation outcomes. The Sama project ended in 2022; no current vendor footprint is established.

Pay & conditions1/3

The October 2026 supplier code sets specific conditions for human-data workers, but it is a written requirement rather than evidence of current implementation. The only cited implementation account is the historical GPT-4-era work; TIME’s 2022 Sama reporting documented worker concerns and OpenAI/Sama disputed or qualified key claims.

Worker voice & remedy1/3

The current code requires channels to pause, decline or escalate unsafe tasks without retaliation, but no public evidence verifies a current feedback channel’s use or outcomes. Historical reporting records workers raising concerns and the vendor/customer responses, not current implementation.

Evidence & follow-through1/3

The code requires suppliers to retain worker and contract records and cooperate with OpenAI or authorized third-party validation on reasonable request, but OpenAI publishes no worker-specific audit results or independent outcomes.

• OpenAI Supplier Code of Conduct ↗
• GPT-4 Technical Report ↗
• OpenAI used Kenyan workers on less than $2 per hour: TIME investigation ↗
• Sama and Content Moderation: What You Need to Know ↗

The March 2026 agreement permits lawful purposes and classified deployment, so this is broad defense enablement; the public agreement does not identify actual targeting, weapons, or combat-intelligence deployments. It bars directing autonomous weapons and domestic surveillance of U.S. persons, including commercial personal data, and disallows high-stakes automated decisions where human approval is required. OpenAI says it retains the safety stack, deploys only through its cloud, uses classifiers and cleared personnel in the loop, and may terminate for breach. These are concrete company-described safeguards, but no independent enforcement outcomes are public, and the March amendment says NSA service would require a separate agreement. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 1 use restrictions + 2 enforcement & oversight + 2 relationship disclosure = 6, capped at 3/10. Known defense ties lower this rating; safeguards can improve it within the relationship’s cap.

The public agreement does not list task orders, deployed model versions, users, operational workflows, audit outcomes, or whether guardrails have been independently tested in use. Contract permission and deployment of a specific model version are not established here.

Use restrictions1/4

The contract explicitly excludes autonomous weapon direction, domestic surveillance of U.S. persons, and specified unapproved high-stakes decisions, but lawful military activity remains authorized and not all harmful surveillance/weapon uses are categorically barred.

Enforcement & oversight2/3

The agreement describes cloud-only deployment, OpenAI-controlled safety stack and classifiers, cleared staff in the loop, and termination rights; these are use-specific contractual and technical review mechanisms, with no independent outcome evidence.

Relationship disclosure2/2

OpenAI publishes the agreement’s scope, restrictions, architecture and NSA carve-out; it still does not disclose deployed mission/use-level details.

A classified DoW contract authorizes broad lawful use, but the public evidence describes no actual targeting, strike, or combat-intelligence support; broad-defense cap applies. The relationship limits this grade to 3/10.

• Our agreement with the Department of War ↗
• Our agreement with the Department of War ↗

OpenAI is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. OpenAI’s April 2026 agreement names Microsoft as its primary cloud partner and a major shareholder, with revenue-share payments to Microsoft through 2030. That is a documented commercial link to a BDS target, not a separate official listing of OpenAI; local use of downloaded weights does not by itself establish a new payment to either company. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.

• BNC guide to boycott and pressure targets ↗
• BNC No Tech campaign ↗
• OpenAI: amended Microsoft partnership, April 27 2026 ↗