Insights→What is Responsible AI : Part 2
LLM

What is Responsible AI : Part 2

Gaurav ChopraGaurav Chopra·November 14, 2025
Building on the foundations established in Part 1, this article examines how industry leaders are implementing responsible AI practices at scale. We explore the frameworks, methodologies, and real-world applications that are setting new standards for ethical AI development and deployment across healthcare, finance, retail, and manufacturing sectors.

1. Anthropic's Constitutional AI

What It Is

Constitutional AI (CAI) represents a groundbreaking method for training AI systems through self-improvement using a constitution—a carefully curated list of principles—without requiring human labels to identify harmful outputs.[1] The approach consists of two distinct phases: Supervised Learning (SL) and Reinforcement Learning from AI Feedback (RLAIF), marking a significant departure from traditional RLHF (Reinforcement Learning from Human Feedback) methods.[2]

How It Works

Phase 1 (Supervised Learning): The model generates self-critiques and revisions of its own responses, then fine-tunes itself on these revised outputs. This self-improvement process allows the AI to identify and correct potentially harmful or biased responses autonomously.[3]

Phase 2 (Reinforcement Learning): The model evaluates pairs of responses based on constitutional principles, trains a preference model from these AI-generated preferences, and then uses RLAIF as a reward signal for further training. This process enables the AI to make judgments about outputs based on explicit ethical guidelines rather than implicit human preferences.[4]

The Constitution

Anthropic's constitution draws from diverse sources including the UN Declaration of Human Rights, trust and safety best practices, DeepMind's Sparrow Principles, Apple's Terms of Service, and principles that encourage non-Western perspectives.[5] This multi-cultural approach ensures the AI system reflects a broad range of ethical viewpoints rather than a singular cultural framework.

Collective Constitutional AI

In a pioneering democratic approach to AI values, Anthropic trained models using public input, gathering 275 principles from thousands of participants through the Polis platform. This "Collective Constitutional AI" experiment involved approximately 1,000 Americans deliberating on rules for AI behavior, casting 38,252 votes and contributing 1,127 statements.[6] The resulting publicly-sourced constitution demonstrates how democratic processes can influence AI development and create more representative AI values.

Key Innovations

  • Transparency: Constitutional AI is more transparent than RLHF because the principles the system follows can be easily inspected and understood, making the AI's decision-making process more legible to users and regulators.[7]
  • Reduced Human Exposure to Harmful Content: By using AI feedback instead of human feedback, CAI dramatically reduces the need for humans to review disturbing or traumatic content during the training process, addressing a significant occupational health concern in AI development.[8]
  • Constitutional Classifiers: The constitution can be used to generate synthetic data for training input/output classifiers that detect and block harmful content, creating an additional layer of safety mechanisms.[9]

2. Google's Responsible AI Practices

Core Framework

Google's AI Principles, established in 2018, serve as a living constitution guiding the development and deployment of AI systems across the organization.[10] These principles are informed by the Secure AI Framework (SAIF), which addresses security and privacy concerns, and the Frontier Safety Framework, which evaluates evolving model capabilities.[11]

The seven foundational principles established in 2018 are:

  1. Be socially beneficial
  2. Avoid creating or reinforcing unfair bias
  3. Be built and tested for safety
  4. Be accountable to people
  5. Incorporate privacy design principles
  6. Uphold high standards of scientific excellence
  7. Be made available for uses that accord with these principles[12]

Governance Process

Google's governance process covers the entire AI lifecycle—from model development through application deployment to post-launch monitoring. The process includes:

  • Risk Identification and Assessment: Conducted through rigorous research, external expert input, and red teaming exercises
  • Evaluation: Systems are evaluated against comprehensive safety, privacy, and security benchmarks
  • Mitigation: Building protections through safety tuning, security controls, and continuous monitoring[13]

Key Commitments

Google has made explicit commitments not to develop AI for:

  • Weapons or technologies that cause overall harm
  • Surveillance that violates internationally accepted norms
  • Purposes that contravene international law and human rights[14]

2024-2025 Updates

Google has implemented a new Frontier Safety Framework in Google DeepMind for evaluating frontier models like Gemini 2.0. This framework includes:

  • Recommendations for Heightened Security: Identifying where the strongest efforts to curb exfiltration risk are needed
  • Deployment Mitigations Procedure: Focusing on preventing misuse of critical capabilities in deployed systems
  • Deceptive Alignment Risk: Addressing the risk of autonomous systems deliberately undermining human control[15]

Demonstrating transparency and accountability, Google has published annual Responsible AI Progress Reports since 2019, providing detailed insights into their governance structures, processes, and outcomes.[16]

3. Microsoft's AI Governance Framework

Six Core Principles

Microsoft's responsible AI approach is built on six foundational principles that guide all AI development and deployment:[17]

  • Fairness: Ensuring AI systems treat all people fairly and avoid creating or amplifying bias
  • Reliability & Safety: Developing systems that perform consistently and safely across diverse conditions
  • Privacy & Security: Protecting user data and maintaining robust security measures
  • Inclusiveness: Empowering everyone regardless of background or ability
  • Transparency: Ensuring stakeholders understand how AI systems work and their limitations
  • Accountability: Maintaining human oversight and responsibility for AI systems

Responsible AI Standard

The Responsible AI Standard serves as Microsoft's internal framework defining product development requirements for responsible AI. This standard consolidates practices to ensure compliance with emerging AI laws and regulations globally.[18] All AI systems undergo Impact Assessments early in development and are subject to appropriate data governance requirements throughout their lifecycle.

Frontier Governance Framework

Introduced in 2025 as a response to the growing complexity of frontier AI models that could pose national security or public safety risks, the Frontier Governance Framework functions as an internal monitoring and risk assessment mechanism for advanced models before release.[19] This framework emerged from voluntary safety commitments made in May 2024 alongside fifteen other leading AI organizations.

Implementation Approach

Microsoft leverages the NIST AI Risk Management Framework, applying its four key functions throughout the AI lifecycle:[20]

  • Govern: Establishing policies, processes, and accountability structures
  • Map: Identifying potential sources of risk and their contexts
  • Measure: Evaluating systems against established benchmarks and metrics
  • Manage: Implementing controls and mitigations to address identified risks

The Sensitive Uses and Emerging Technologies program provides pre-deployment review for high-impact and higher-risk AI applications. In 2024, 77% of consultations through this program were for generative AI systems, demonstrating the program's critical role in responsible deployment.[21]

Red Teaming

Microsoft's AI Red Team conducted 67 operations in 2024 across flagship models including the Phi series and Copilot tools, stress-testing them for vulnerabilities to malicious prompts and potential misuse.[22] These red teaming exercises are paired with automated measurement pipelines designed to simulate adversarial interactions and flag harmful content generation before public release, now covering multiple modalities including text, audio, and video.

Customer Support

Microsoft offers comprehensive tools and resources to help customers build AI responsibly, including:

  • Responsible AI Dashboard: Tools for monitoring and managing AI systems
  • Transparency Notes: Detailed documentation on how AI systems work
  • Technical Tools: Including fairness assessment components and model interpretability features[23]

4. Real-World Case Studies

Healthcare Success: Smart Impression (Microsoft/Nuance)

Industry: Healthcare - Radiology

The Challenge

The radiology profession faces a critical workforce crisis. More than 54% of radiologists report feelings of burnout (up from 49% in 2022), with up to 44% reporting that burnout has strong or severe impacts on their lives. Vacancy rates have reached 18%—a 20-year high—while case volume and complexity continue to increase.[24]

The Solution

PowerScribe Smart Impression, built on the PowerScribe platform used by more than 80% of all radiologists, is an AI-powered productivity tool that streamlines radiology workflows and accelerates reporting. The system automatically creates draft impressions and recommendations using advanced generative AI, processing information from the protocol name and radiologist-dictated findings section.[25]

Responsible AI Implementation

Through Microsoft's Sensitive Uses review process, the product team identified and mitigated key risks related to using AI in healthcare settings. The system underwent rigorous testing with 9.8 million radiology reports from multiple sites to ensure accuracy and reliability.[26] Three radiologists assessed 3,879 reports across multiple imaging modalities from 8 U.S. imaging sites to evaluate the system's performance.

Results and Impact

  • Efficiency Gains: Three-quarters of radiologists using the system report improved efficiency, with some saving up to a minute per read—quickly adding up over 30-40 cases each day[27]
  • Quality Improvement: The system helps radiologists avoid omissions and identifies possible misspellings or errors, contributing to higher reporting quality
  • Reduced Cognitive Burden: Natural, accurate wording of draft impressions reduces mental fatigue, particularly toward the end of shifts
  • Work Satisfaction: Radiologists report increased satisfaction at work, with the tool helping them "not forget things" in their impressions[28]

Financial Services: Wells Fargo

Industry: Financial Services - Banking

The Challenge

Wells Fargo's 35,000 bankers across 4,000 branches needed instant access to guidance on 1,700 internal procedures. The traditional process of seeking information from colleagues or searching through documentation was time-consuming and inefficient, with average response times of 10 minutes for basic queries.

The Solution

Wells Fargo developed a Microsoft Teams app integrated with advanced language models, providing employees with immediate access to procedural information. The app enables bankers to quickly find the information they need without interrupting colleagues or spending time searching through documentation.[29]

Responsible AI Implementation

The bank followed a careful approach to AI deployment, emphasizing thorough testing and gradual rollout. Wells Fargo's CIO emphasized the importance of "not getting caught up with shiny objects" and being "very thoughtful" about AI implementation.[30] The system was designed with strong privacy controls to ensure customer data security and compliance with financial regulations.

Results and Impact

  • Dramatic Efficiency Improvement: 75% of searches are now conducted through the app, reducing response times from 10 minutes to just 30 seconds[31]
  • Widespread Adoption: The tool is deployed to 35,000 bankers across the organization
  • Enhanced Productivity: Employees can quickly resolve customer queries without delays
  • AI Virtual Assistant Expansion: Wells Fargo created an additional AI virtual assistant tool to help Treasury sales teams proactively engage in relevant conversations with customers

Additionally, Wells Fargo's AI-powered virtual assistant Fargo has achieved remarkable success, handling more than 200 million customer requests without sending customer data to language models, demonstrating strong privacy protection in practice.[32]

Retail: CarMax

Industry: Automotive Retail

The Challenge

As the United States' leading used car retailer with more than 45,000 cars available at any given time, CarMax needed to help customers conduct pre-purchase research effectively. With thousands of customer reviews for popular models, potential buyers were overwhelmed by the volume of information. Creating manual summaries for approximately 5,000 car make-model-year combinations would have taken an estimated 11 years using traditional content creation methods.[33]

The Solution

CarMax partnered with OpenAI through Microsoft Azure OpenAI Service to leverage GPT-3 natural language models. The system was designed to abstractly summarize and fine-tune 100,000 customer reviews into 5,000 well-written, digestible summaries that surface key takeaways for each make, model, and year of vehicle in CarMax's inventory.[34]

Responsible AI Implementation

CarMax implemented several responsible AI practices:

  • Human Review: After AI generates content, CarMax staff members review the text to ensure it makes sense in context and fits with the CarMax brand voice[35]
  • Azure's Enterprise Capabilities: Leveraging Azure's security, compliance, reliability, and other enterprise-grade capabilities to protect customer data
  • Quality Control: The system achieved an 80% editorial review approval rate after fine-tuning, demonstrating high-quality output

Results and Impact

  • Accelerated Timeline: CarMax completed in just a few months what would have taken 11 years manually—a 100x improvement in speed[36]
  • Scale Achievement: Successfully created 5,000 unique summaries from 100,000 customer reviews
  • SEO Benefits: AI-generated content boosted search engine rankings and page views
  • Customer Value: Shoppers can now quickly understand key vehicle characteristics—whether it's a great family car, how comfortable the ride is, or if there's enough space for weekend adventures
  • Business Growth: Increased customer traffic drove substantial business growth, with the investment paying for itself many times over

According to Ritu Jyoti, Group Vice President of AI and Automation Research at IDC: "CarMax was at the forefront of embracing generative AI responsibly in partnership with Microsoft and has been a successful industry disrupter that has transformed the process of buying a used car."[37]

Manufacturing: Shell

Industry: Energy - Manufacturing and Production

The Challenge

Shell operates thousands of critical pieces of equipment across its global upstream, downstream, and integrated gas assets. Traditional reactive maintenance approaches led to costly unplanned downtime, production interruptions, and potential safety and environmental risks. The company needed a way to predict equipment failures before they occurred while managing the complexity of a vast, globally distributed asset base.

The Solution

Shell deployed AI-driven predictive maintenance at unprecedented scale, powered by C3 AI technology running on Microsoft Azure. The system monitors more than 10,000 pieces of equipment including control valves, pumps, compressors, and other critical components across Shell's global operations.[38]

Technical Infrastructure

The scale of Shell's implementation is remarkable:

  • Data Volume: Ingests 20 billion rows of data weekly from more than 3 million data streams[39]
  • Machine Learning Models: Trains, tunes, and runs nearly 11,000 production-grade ML models
  • Predictions: Generates over 15 million predictions daily
  • Global Reach: Supports deployment across Shell's upstream, downstream, and integrated gas assets worldwide[40]

Responsible AI Implementation

According to Dan Jeavons, Vice President of Computational Science & Digital Innovation at Shell: "Monitoring 10,000 pieces of critical equipment with AI-enabled predictive maintenance is an important milestone for Shell—an ambitious target we had set for 2021 and successfully achieved. We have an exceptionally talented team to thank for this accomplishment, as well as partners like C3 AI, whose technology helped us reach this level of scale."[41]

Results and Impact

  • Proactive Maintenance: The system identifies equipment degradation and potential failures before they occur with very high levels of precision[42]
  • Cost Savings: In one documented example, alerts enabled proactive repair of almost 90 control valves, saving millions of dollars in avoided downtime and repair costs. Broader analyses indicate a 20% decrease in overall maintenance costs (approximately $2 billion annually)[43]
  • Operational Efficiency: 35% reduction in unplanned asset downtime
  • Safety Improvements: 40% reduction in safety incidents related to equipment failure
  • Environmental Benefits: Preventing equipment failures reduces potential environmental risks and emissions

Shell is now exploring extending these AI capabilities to renewable energy assets including wind farms and solar parks, as well as other use cases such as production optimization, industrial process design, and sustainability solutions.[44]

Generative AI: Adobe Firefly

Industry: Creative Technology - Generative AI

The Challenge

Marketing and creative leaders recognized generative AI's potential for content creation and personalization, but faced significant concerns about copyright infringement, intellectual property violations, and brand safety. Many AI tools were trained on datasets scraped from the internet without proper licensing, creating legal and ethical risks for businesses using AI-generated content.

The Solution

Adobe Firefly was designed from the ground up to be commercially safe, built with transparency and responsible AI practices at its core. Unlike many generative AI tools, Firefly is trained exclusively on content where Adobe has permission or rights.[45]

Responsible AI Implementation

Training Data Transparency: Firefly's first commercial model was trained on:

  • Licensed content from Adobe Stock
  • Public domain content where copyright has expired
  • Openly licensed content[46]

Key Protections:

  • No Customer Data Training: Adobe does not train Firefly models on Creative Cloud or Adobe Experience Cloud subscribers' personal content[47]
  • IP Indemnification: Customers on qualifying plans are eligible for IP indemnification for generated content, with Adobe offering to pay legal bills for enterprise users sued for copyright violations[48]
  • Content Credentials: As a founding collaborator of the Content Authenticity Initiative (CAI), Adobe automatically attaches Content Credentials to Firefly-generated images, functioning as a digital "nutrition label" showing how and when content was created[49]
  • Custom Models: Enterprise users can deploy Firefly Custom Models trained on approved brand assets, ensuring on-brand content creation while maintaining IP safety

Results and Impact

  • Enterprise Confidence: Commercial-safe generation through transparent training data reassured enterprise customers about copyright and licensing concerns
  • Brand Safety: Major brands like Dentsu, Gatorade, and Stagwell are testing Firefly, signaling wider enterprise adoption[50]
  • Creative Acceleration: Teams can create thousands of assets without manual copyright checks, empowering junior team members to generate content confidently
  • Competitive Advantage: Firefly's "IP-safe AI available in industry-standard tools" approach helps gain widespread adoption in copyright-conscious creative industries[51]

According to eMarketer's Gadjo Sevilla: "Major brands like Dentsu, Gatorade, and Stagwell are already testing Firefly, signaling wider enterprise adoption. Making IP-safe AI available in industry-standard tools can help Firefly, and by extension Adobe, gain widespread adoption in copyright-friendly AI image generation."[52]

Contributor Support: Adobe has established a Firefly Contributor bonus program for Adobe Stock contributors whose content was used in training, demonstrating commitment to supporting the creative community.[53]

5. Key Success Patterns Across Industries

Analysis of these implementations reveals common success factors that organizations should consider when deploying responsible AI:

Common Success Factors

  1. Strong Governance Frameworks: All industry leaders have formal, documented approaches to AI governance with clear principles, standards, and accountability structures
  2. Continuous Monitoring: Real-time tracking of AI behavior, performance metrics, and potential risks enables rapid identification and mitigation of issues
  3. Human Oversight: Maintaining human-in-the-loop for critical decisions ensures accountability and appropriate intervention when needed
  4. Transparency: Clear documentation, explainability features, and open communication about AI capabilities and limitations build trust with users and stakeholders
  5. Training & Education: Comprehensive programs for employees ensure proper understanding and use of AI systems, from technical teams to end users
  6. Cross-functional Teams: Bringing together ethics experts, legal counsel, technical specialists, and business stakeholders ensures diverse perspectives and comprehensive risk assessment

Industry-Specific Insights

Healthcare: Emphasis on clinical validation, patient safety, and reducing clinician burnout while maintaining quality of care

Financial Services: Focus on regulatory compliance, customer data privacy, and fraud detection while improving customer experience

Retail: Balancing personalization with privacy, ensuring transparency in AI-driven recommendations and pricing

Manufacturing: Prioritizing safety, environmental protection, and operational efficiency through predictive analytics

Creative Industries: Addressing copyright concerns, supporting creative communities, and ensuring commercial viability of AI-generated content

Conclusion

The case studies and frameworks examined in this article demonstrate that responsible AI is not just an ethical imperative—it's a business necessity. Organizations that prioritize responsible AI practices from the outset:

  • Build trust with customers, regulators, and stakeholders
  • Reduce legal and reputational risks
  • Achieve better adoption and user satisfaction
  • Create sustainable competitive advantages
  • Drive meaningful business value while minimizing harm

As AI continues to evolve and permeate every aspect of business and society, the approaches pioneered by Anthropic, Google, Microsoft, and the organizations profiled in these case studies provide a roadmap for others to follow. The key is not to view responsible AI as a constraint on innovation, but rather as an enabler of sustainable, ethical, and impactful AI deployment at scale.

Success in responsible AI requires ongoing commitment, continuous learning, and willingness to adapt as technology and societal expectations evolve. Organizations that embrace these principles today will be best positioned to harness AI's transformative potential while building a more equitable and beneficial future for all.

References

  1. Anthropic. (2022). "Constitutional AI: Harmlessness from AI Feedback." Retrieved from: https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback
  2. Bai, Y., et al. (2022). "Constitutional AI: Harmlessness from AI Feedback." arXiv:2212.08073. Retrieved from: https://arxiv.org/abs/2212.08073
  3. Anthropic. (2022). "Constitutional AI: Harmlessness from AI Feedback - Technical Paper." Retrieved from: https://arxiv.org/pdf/2212.08073
  4. Lambert, N. "Constitutional AI & AI Feedback." RLHF Book. Retrieved from: https://rlhfbook.com/c/13-cai.html
  5. Anthropic. (2023). "Claude's Constitution." Retrieved from: https://www.anthropic.com/news/claudes-constitution
  6. Anthropic. (2023). "Collective Constitutional AI: Aligning a Language Model with Public Input." Retrieved from: https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language-model-with-public-input
  7. Anthropic. (2022). "Constitutional AI: Technical Overview." Anthropic ConstitutionalAI v2 PDF.
  8. Generative AI. (2024). "Claude AI's Constitutional Framework: A Technical Guide to Constitutional AI." Medium. Retrieved from: https://medium.com/@genai.works/
  9. Anthropic. (2022). "Constitutional AI Paper - GitHub Repository." Retrieved from: https://github.com/anthropics/ConstitutionalHarmlessnessPaper
  10. Google. (2025). "AI Principles." Retrieved from: https://ai.google/principles/
  11. Google. (2025). "Responsible AI | Google Cloud." Retrieved from: https://cloud.google.com/responsible-ai
  12. Pichai, S. (2018). "AI at Google: our principles." Google Blog. Retrieved from: https://blog.google/technology/ai/ai-principles/
  13. Google AI. "AI Principles - Responsibility." Retrieved from: https://ai.google/responsibility/principles/
  14. Google. "Responsible AI | Google Public Policy." Retrieved from: https://publicpolicy.google/responsible-ai/
  15. Google. (2025). "Responsible AI: Our 2024 report and ongoing work." Google Blog. Retrieved from: https://blog.google/technology/ai/responsible-ai-2024-report-ongoing-work/
  16. Google. (2023). "Responsible AI: Looking back at 2022, and to the future." Google Blog. Retrieved from: https://blog.google/technology/ai/responsible-ai-looking-back-at-2022-and-to-the-future/
  17. Microsoft. "Responsible AI Principles and Approach." Retrieved from: https://www.microsoft.com/en-us/ai/principles-and-approach/
  18. Microsoft. "What is Responsible AI - Azure Machine Learning." Retrieved from: https://learn.microsoft.com/en-us/azure/machine-learning/concept-responsible-ai
  19. AI Magazine. (2025). "Inside Microsoft's 2025 Responsible AI Transparency Report." Retrieved from: https://aimagazine.com/articles/inside-microsofts-2025-responsible-ai-transparency-report
  20. Ningegowda, K. (2025). "Microsoft's 2025 AI Transparency Report: Evolving Responsible AI Governance Beyond Fairness." Medium. Retrieved from: https://medium.com/@keerthi.ningegowda/microsofts-2025-ai-transparency-report
  21. Microsoft. (2023). "Putting principles into practice: How we approach responsible AI at Microsoft." Microsoft PDF Document.
  22. Technology Magazine. (2025). "How Microsoft is Making AI R&D Responsible & Transparent." Retrieved from: https://technologymagazine.com/articles/inside-microsofts-2025-responsible-ai-transparency-report
  23. Microsoft. "Responsible AI: Ethical policies and practices." Retrieved from: https://www.microsoft.com/en-us/ai/responsible-ai
  24. Nuance. (2023). "Nuance Accelerates the Adoption of Generative AI for Radiology with PowerScribe Advanced Auto Impression Capability." Retrieved from: https://news.nuance.com/2023-11-28-Nuance-Accelerates-the-Adoption-of-Generative-AI-for-Radiology
  25. Microsoft. (2023). "How Microsoft and Nuance empower radiologists with AI-powered solutions." Microsoft Industry Blogs. Retrieved from: https://www.microsoft.com/en-us/industry/blog/healthcare/2023/12/01/
  26. Doshi, S., et al. (2024). "Artificial intelligence-generated smart impression from 9.8-million radiology reports." medRxiv. Retrieved from: https://www.medrxiv.org/content/10.1101/2024.03.07.24303787v1.full
  27. Microsoft. (2024). "RSNA 2024: AI's impact inside and outside the reading room." Microsoft Industry Blogs. Retrieved from: https://www.microsoft.com/en-us/industry/blog/healthcare/2024/12/11/rsna-2024
  28. Microsoft. "PowerScribe One | Microsoft for Healthcare." Retrieved from: https://www.microsoft.com/en-us/health-solutions/radiology-workflow/powerscribe-one
  29. Microsoft. (2025). "AI-powered success—with more than 1,000 stories of customer transformation and innovation." The Microsoft Cloud Blog. Retrieved from: https://www.microsoft.com/en-us/microsoft-cloud/blog/2025/07/24/ai-powered-success
  30. AIM Research. (2025). "Fargo Handled More Than 200 Million Requests Without Sending Customer Data to an LLM." Retrieved from: https://aimmediahouse.com/market-industry/fargo
  31. ServerSys. (2025). "AI in Action: Insights from Microsoft's Work Trends Index 2025." Retrieved from: https://www.serversys.com/insights/
  32. American Banker. (2025). "Wells Fargo plans to unleash Google's agentic AI bank-wide." Retrieved from: https://www.americanbanker.com/news/wells-fargo
  33. Microsoft. (2023). "Build: Azure OpenAI Service helps customers accelerate innovation with large AI models." Microsoft Source. Retrieved from: https://news.microsoft.com/source/features/ai/azure-openai-service
  34. Microsoft. "CarMax puts customers first with car research tools powered by Azure OpenAI Service." Microsoft Customer Stories. Retrieved from: https://www.microsoft.com/en/customers/story/carmax
  35. Prismware. (2024). "CarMax puts customers first with car research tools powered by Azure OpenAI Service." Retrieved from: https://www.prismware.co/carmax
  36. Voicebot.ai. (2022). "Microsoft Enhances Azure OpenAI Service and Widens Access to GPT-3 Models." Retrieved from: https://voicebot.ai/2022/05/25/microsoft-enhances-azure-openai-service/
  37. Rooney, P. (2025). "CarMax drives business value with GPT-3.5." CIO. Retrieved from: https://www.cio.com/article/475487/carmax-drives-business-value-with-gpt-3-5.html
  38. C3 AI. (2025). "The Scale of Shell's Global AI Predictive Maintenance Program." Retrieved from: https://c3.ai/blog/how-shell-scaled-ai-predictive-maintenance
  39. Reliabilityweb. (2022). "Shell Scales Artificial Intelligence Fueled Predictive Maintenance to 10,000 Assets." Retrieved from: https://reliabilityweb.com/en/press-release/
  40. C3 AI. (2022). "Shell Achieves Major Milestone: Scales Artificial Intelligence Predictive Maintenance to 10,000 Pieces of Equipment Using C3 AI." Business Wire. Retrieved from: https://www.businesswire.com/news/home/20220308005567/en/
  41. SPE Journal of Petroleum Technology. (2022). "Shell Scales Predictive Maintenance to 10,000 Pieces." Retrieved from: https://jpt.spe.org/shell-scales-predictive-maintenance
  42. C3 AI Investor Relations. (2022). "Shell Achieves Major Milestone." Retrieved from: https://ir.c3.ai/news-releases/
  43. Klover.ai. (2025). "Shell's AI Strategy: Analysis of AI Dominance in Energy." Retrieved from: https://www.klover.ai/shell-ai-strategy-analysis
  44. Decarbonfuse. (2022). "Shell Achieves Major Milestone: Scales Artificial Intelligence Predictive Maintenance." Retrieved from: https://decarbonfuse.com/posts/shell-achieves-major-milestone
  45. Adobe. (2025). "Adobe Firefly | Comprehensive & Commercially Safe AI Content Creation for Businesses." Retrieved from: https://business.adobe.com/products/firefly-business/firefly-ai-approach.html
  46. Adobe. (2025). "Adobe Firefly FAQ." Retrieved from: https://helpx.adobe.com/firefly/web/get-started/
  47. Adobe. "Adobe Firefly - Free Generative AI for creatives." Retrieved from: https://www.adobe.com/products/firefly.html
  48. Nover, S. (2025). "Adobe's Firefly is impressive and promises it's copyright-safe." GZERO Media. Retrieved from: https://www.gzeromedia.com/gzero-ai/adobes-firefly
  49. Adobe Blog. (2024). "Growing responsibly in the age of AI: Adobe Firefly and Stock." Retrieved from: https://blog.adobe.com/en/publish/2024/04/16/growing-responsibly
  50. Adobe. (2025). "Adobe Firefly for Enterprise | Generative AI for Content Creation." Retrieved from: https://business.adobe.com/products/firefly-business.html
  51. FUD. (2025). "Are You Sure Your Brand Content Is Truly Safe? Discover How Adobe Firefly Ensures Commercial-Safe Content Creation." Retrieved from: https://blog.joinfud.com/p/
  52. Overdrive. (2025). "How to Brand-Safe with AI-Generated Images: Why Getty and Firefly Lead the Way." Retrieved from: https://www.ovrdrv.com/insights/
  53. Adobe. (2025). "Firefly FAQ for Adobe Stock Contributors." Retrieved from: https://helpx.adobe.com/stock/contributor/help/firefly-faq
Gaurav Chopra
Gaurav Chopra

Gaurav is a Co-Founder of Eightgen AI

Work with us

Found this useful? Let's talk about your build.

We write about what we build. If any of this resonates with a challenge you're facing, book a free 30-minute call — no prep needed.

Book a free call →contact@eightgen.ai