Artificial Intelligence 12 min

How to Regulate AI Without Stifling Innovation

Responsible AI innovation doesn't have to mean choosing between progress and protection. Members of the Senior Executive AI Think Tank share 10 policy recommendations for building trust, strengthening accountability and protecting people without putting the brakes on innovation.

by AI Editorial Team on November 18, 2026

Every day, AI makes decisions, handles sensitive information and takes on responsibilities that once belonged exclusively to people. But when an algorithm gets it wrong, who answers for the consequences? Who ensures confidential business data stays confidential? And how can companies be held accountable without creating regulations that favor tech giants and leave smaller innovators behind?

These aren’t hypothetical questions. They’re the challenges business leaders are already confronting as AI moves from experimentation into everyday operations. The debate over AI regulation isn’t just about whether the technology needs guardrails, but about deciding where those guardrails belong, who should be responsible for enforcing them and how to protect people without putting the brakes on progress.

We asked members of the Senior Executive AI Think Tank, a curated group of experts in machine learning, generative AI and enterprise AI applications, to weigh in on the one law, regulation or policy they believe is essential to responsible AI development and deployment. Their perspectives explore what effective oversight should look like, where regulatory responsibility belongs and how policymakers can strike the right balance between protecting the public and giving the AI industry room to grow.

“You cannot govern what you cannot see, and right now regulators and the industry are both flying blind. ”

Sarah Choudhary, CEO of Ice Innovations

– Sarah Choudhary, CEO of Ice Innovations

SHARE IT

Make AI Incident Reporting a Federal Requirement

For Sarah Choudhary, CEO of Ice Innovations, one of the most important regulatory priorities is establishing a federal AI incident reporting law modeled on aviation safety practices. Her recommendation addresses a fundamental weakness in the current AI ecosystem: Organizations often resolve failures internally without sharing what happened or what others could learn from the experience.

“When a plane has a near miss, the pilot reports it and the whole industry learns. AI has no such habit,” Choudhary says.

She believes reporting requirements should extend beyond advanced AI developers to organizations deploying AI in consequential areas such as hiring, lending, healthcare and safety. Companies should report serious failures within days, share lessons without identifying themselves publicly and receive protection when reporting incidents in good faith.

“This does not tell anyone what to build. It tells them to say what broke,” she explains. “One national rule also beats fifty state versions, which is the real tax on innovation.”

A consistent federal reporting framework could help organizations learn from failures without dictating how they develop their products. It also aligns with the European Union AI Act’s serious incident reporting requirements, which establish reporting obligations for serious incidents involving high-risk AI systems.

“You cannot govern what you cannot see, and right now regulators and the industry are both flying blind.”

Make Accountability the Foundation of AI Trust

Rodney Mason, Chief Marketing Officer at Minty, advocates for a policy that makes accountability the foundation of responsible AI adoption. Rather than treating regulation as a barrier to technological progress, he believes businesses should demonstrate that they can be trusted to manage the outcomes their systems produce.

“AI should earn trust through accountability, not be constrained by fear of innovation,” Mason says. “Require organizations to take responsibility for the outcomes their AI creates, with transparency, appropriate safeguards and human accountability where it matters.”

His recommendation emphasizes regulating the impact of AI rather than prescribing how companies must build it. That gives developers room to experiment with different models, tools and approaches while ensuring organizations remain accountable for their applications.

“The goal should be to make trust an accelerator of innovation, not a barrier to it,” he adds.

Establishing ownership, documenting safeguards and communicating how AI influences decisions can help organizations build confidence among employees, customers and business partners.

“Data training should require explicit consent, and not vague permissions buried in the fine print and subject to change.”

Korena Keys, Founder and Visionary of KeyMedia Solutions

– Korena Keys, Founder and Visionary of KeyMedia Solutions

SHARE IT

Protect Confidential Business Data Used by AI

Korena Keys, Founder and Visionary of KeyMedia Solutions, identifies confidential data protection as a critical requirement for responsible AI deployment.

“I believe we need a law that protects confidential information shared with AI tools (specifically paid subscriptions) and how that information is then used,” Keys says. 

She illustrates the issue with a company uploading financial information for a five-year trend analysis. That information, she argues, should not automatically become part of a broader AI training dataset or potentially accessible to outside parties.

“Data training should require explicit consent, and not vague permissions buried in the fine print and subject to change,” she explains.

Keys also calls for clear data-use disclosures, safeguards against unauthorized access and enforceable retention and deletion requirements.

Organizations should understand their providers’ data-handling practices, establish approved-use policies and verify whether confidential inputs can be retained or used for model training.

“We would all benefit from using AI tools without fear of surrendering confidential data,” Keys says.

Protecting business information through enforceable standards could make organizations more comfortable adopting AI while preserving opportunities to explore new applications.

Tie Independent Testing to the Potential for Harm

Egbert von Frankenberg, CEO of Knightfox App Design Ltd., recommends a risk-tiered regulatory framework requiring independent testing of high-stakes AI systems before deployment and incident reporting afterward.

“High-stakes means decisions about credit, hiring, health, safety or critical infrastructure; everything else stays lightly regulated or exempt,” von Frankenberg says.

He argues that developers should not be solely responsible for evaluating their own systems. Bias, unsafe outputs and security weaknesses can remain undetected when organizations lack independent scrutiny.

“Builders grading their own work have weak incentives to look hard,” he says. “Incident reporting, like aviation’s, lets the whole industry learn from failures.”

The recommendation is consistent with the risk-based approach of the NIST AI Risk Management Framework, which emphasizes identifying, assessing and managing risks throughout an AI system’s lifecycle.

For executives, the takeaway is to assess applications according to their potential consequences, not simply the sophistication of the underlying technology or the size of the company deploying it.

Build Shared Testing Standards Without Creating Barriers to Entry

Lynn Comp, Head of AI Center of Excellence at Intel, proposes an open industry test bed for model safety rather than regulations primarily shaped by large technology companies. Her concern is that excessive regulatory influence from established players could make compliance more difficult for startups and emerging competitors.

“Regulatory capture as a moat will stagnate the entire industry,” Comp says.

She also emphasizes that AI safety cannot be separated from cybersecurity. Existing protections and penalties for cyberattacks should extend to model vendors when their products introduce significant security risks.

“If the tool is so high risk, a legal requirement for test and validation methodology and penalties for sloppy or poor guardrails on basic security practice fundamentals are a must-do,” she explains.

Comp argues that organizations should involve security experts in model releases and apply established software safety practices rather than relying exclusively on specialized controls designed for large language models.

Her proposal offers a practical middle ground: Shared testing resources, transparent evaluation methodologies and enforceable security expectations could improve industry-wide safety without requiring every organization to build expensive testing infrastructure independently.

Require Traceability Throughout the AI Lifecycle

Aishwarya Shah, Independent Researcher, recommends a mandatory AI Accountability and Traceability Framework proportionate to the risks associated with each application.

“Organizations deploying high-impact AI should be able to demonstrate where critical data originated, how decisions were reached, who owns the outcomes and what safeguards exist when systems fail,” Shah says.

She emphasizes that accountability should not translate into identical requirements for every AI application.

“Low-risk experimentation needs flexibility, while systems influencing financial, healthcare, employment or other consequential decisions require stronger oversight,” she explains.

Organizations need enough documentation to understand and explain their systems without creating unnecessary administrative burdens for routine applications.

“Trust should be demonstrated through evidence, not assumed through policy statements,” she says.

For executives, that means establishing traceability as an ongoing management responsibility, with documented data sources, decision processes and accountability throughout deployment.

“If AI screens out a qualified applicant because it misreads an industrial certification, the applicant should know AI was involved, see the reason and reach a human who can reverse the decision.”

Fabio Danze Montini, Investor and Owner of FDM Industrial Sales & Marketing SL

– Fabio Danze Montini, Investor and Owner of FDM Industrial Sales & Marketing SL

SHARE IT

Give Individuals the Right to Challenge AI Decisions

Fabio Danze Montini, Investor and Owner of FDM Industrial Sales & Marketing SL, addresses a practical problem: Even a sophisticated AI system can misunderstand information or make an incorrect assessment that materially affects someone’s life.

“I’d require any company that deploys AI to make a decision about someone’s job, credit or access to essential services to give that person a right to appeal,” Montini says.

He illustrates the issue with an industrial job applicant whose qualifications are incorrectly assessed by an AI screening system.

“If AI screens out a qualified applicant because it misreads an industrial certification, the applicant should know AI was involved, see the reason and reach a human who can reverse the decision,” he explains.

Importantly, he believes responsibility should remain with the organization using the system, even when the technology comes from an outside vendor.

“Tying the rule to the consequences of a decision leaves teams free to experiment with low-risk uses,” Montini says.

For business leaders, this means establishing clear escalation procedures and ensuring human reviewers have both the authority and information needed to correct mistakes.

Make Human Review an Enforceable Right

Dr. Aditya Vikram Kashyap, Vice President of Firmwide Innovation at Morgan Stanley, similarly advocates for a legal right to appeal decisions materially influenced by AI.

His proposal goes further by specifying what organizations must provide.

“The institution issuing the decision would have to disclose how AI influenced it, identify the principal reasons for the denial and offer timely human review by someone with authority to reverse it,” Dr. Kashyap says.

He stresses that outsourcing AI technology should not transfer responsibility away from the institution making the decision.

“The appeal would be free to the person affected,” he explains. “For a mortgage applicant, that means being able to challenge an incorrect income figure and get a decision in time to act on it.”

Dr. Kashyap also believes regulators should be able to require organizations to reopen affected cases and impose penalties when appeal rights are denied.

At the same time, he emphasizes that this policy would not prevent companies from developing, testing or comparing competing AI models before deployment.

Require Transparency When High-Impact Models Change

Vivek Kumkar, Sr. GenAI Leader at Amazon Web Services (AWS), highlights an often-overlooked regulatory challenge: AI systems change continuously.

Model updates can improve accuracy, introduce new capabilities and strengthen safeguards. However, they can also change how a system behaves after an organization has completed its original testing.

Kumkar recommends requiring providers to issue change notices when updating models used in lending, hiring and healthcare decisions.

“Model updates are how most AI systems become more accurate and safer, so regulation should make frequent updates easy to trust rather than slow to ship,” he says. “Before an update reaches those customers, the provider should state what changed, such as new weights or safety filters, and how it was tested.”

This information would allow deploying organizations to evaluate changes against their own use cases before switching to a new version. It would also establish a clear reference point for audits and appeals.

Kumkar points to existing regulatory approaches for comparison, including the FDA’s authorization of predetermined change plans for certain AI medical devices and the European Union’s requirements for reassessing substantially modified high-risk systems.

“A change notice would extend that discipline to deployers at little cost and let providers keep improving their models without waiting for approval.”

Enforce Existing Laws and Hold Companies Accountable

Peter Boyd, General Guru and CEO of PaperStreet Web Design, takes a different position. Rather than advocating for another comprehensive AI-specific law, he believes existing legal protections should be enforced more consistently.

“There are already laws on the books that will force companies to implement AI responsibly,” Boyd says. “What we need is state attorneys general and plaintiff lawyers to hold companies accountable for their misdeeds.”

He argues that organizations should face meaningful consequences when they release technology that causes damage.

“If someone releases code into the wild that causes economic or physical damage, then criminal or civil charges should be brought,” he explains.

Boyd believes a prominent enforcement action could influence industry behavior more effectively than additional rules alone.

“It only takes one high-profile case to compel everyone to put in better safeguards,” he says.

10 Practical Steps to Make AI More Accountable

  • Establish an AI incident reporting process. Create internal procedures for documenting serious failures and near misses, with clear escalation timelines and a commitment to sharing lessons across the organization.
  • Make accountability part of AI adoption. Assign responsibility for AI outcomes, establish appropriate safeguards and ensure human oversight where decisions have significant consequences.
  • Protect confidential information. Review AI vendor data policies, require explicit consent for training uses and establish clear retention, access and deletion requirements.
  • Introduce independent testing for high-risk applications. Evaluate systems for bias, security vulnerabilities and unsafe outputs before deployment, prioritizing applications with the greatest potential for harm.
  • Support open and consistent safety testing. Participate in industry testing initiatives and establish security validation practices that do not create unnecessary barriers for smaller competitors.
  • Document AI decisions and data sources. Maintain records showing where important data originated, how systems influence decisions and who is responsible for addressing problems.
  • Create accessible appeal mechanisms. Give individuals affected by consequential AI decisions a clear explanation and access to a qualified human reviewer who can reverse mistakes.
  • Make human review an enforceable right. Ensure individuals can challenge consequential AI-driven decisions through a timely, free appeal process with meaningful human oversight.
  • Implement model change management. Require providers to communicate significant updates and give internal teams an opportunity to retest high-impact applications before adopting new versions.
  • Enforce existing laws and hold companies accountable. Review existing legal obligations, establish clear accountability for AI-related harm and ensure organizations face meaningful consequences when their technology causes damage.

The Future of AI Depends on Getting Accountability Right

The biggest risk to AI innovation may not be regulation itself, but getting regulation wrong. Rules that are too broad could entrench the industry’s largest players, discourage experimentation and make it harder for emerging companies to compete. Rules that are too weak could undermine public confidence and leave businesses exposed to consequences they aren’t prepared to manage.

Business leaders have an important role in shaping an AI ecosystem where innovation and accountability reinforce one another. Organizations that invest in responsible practices now will be better prepared for evolving regulations, more likely to earn stakeholder trust and better positioned to make AI adoption sustainable. Ultimately, the strongest foundation for AI’s future is an industry where protecting people and building better technology go hand in hand.


Copied to clipboard.