Americans Are Worried About AI Safety. The State of Corporate Disclosure Isn’t Helping.
Our latest research reveals a wide gap between the public’s concerns about critical AI safety, oversight, and governance issues and the level of public disclosure among 1,002 of the largest US-based companies on how these issues are being addressed. A large majority of AI users support a slowdown on AI development, but say it will help build trust in the models.
“AI developers and their enterprise customers are becoming increasingly aware that earning the public’s trust on AI is the key to success,” said Just Capital CEO Martin Whittaker. “This study adds to the body of research we’ve been building to help business leaders understand how to go about doing that.”
Key Findings
The strongest support for slowing down on AI comes from the heaviest users
78% of Americans who use AI every day support pausing or slowing AI development so each model release can be evaluated more carefully, compared with 72% of Americans overall.
72% of daily users say a slowdown would increase their trust in AI models, well above the 55% of Americans overall who say the same.
Daily users are also the most attuned to the costs of slowing down. 46% say more competition from Chinese AI models would be a major problem if AI development paused, against 33% of Americans overall, and 45% say the same about the US losing technological or military dominance, against 40% overall.
The issues of greatest concern center on safety and security
Among Americans who support a pause, the concerns they would most want addressed are data privacy and security (21%) and AI safety, the risk of losing control, or AI causing harm (20%), followed by the accuracy and reliability of AI outputs (16%).
Job loss and economic displacement (10%) and broader concerns about societal overreliance on AI (10%) trail behind. Nearly a quarter (23%) don’t name a specific concern.
Nearly half of America’s largest companies have yet to formally disclose how they’re approaching critical AI safety, governance and oversight issues
In a review of 1,002 of the largest US-based companies including eight hyperscalers across nine disclosure areas spanning AI governance, data protection, workforce impacts, and community initiatives, disclosure is limited.
Nearly half of 1,002 companies reviewed (498) disclose on none of nine AI-related issues assessed. 48 of those companies are in the Technology industry.
Only 10% of companies disclose a policy or set of principles that explicitly addresses preventing harm, deception, or manipulation; 14% disclose a policy or set of principles that specifically mentions keeping humans in charge, human oversight, or maintaining human-in-the-loop; 23% disclose a policy or set of principles that specifically mentions protecting personal data or privacy
Salesforce is the only company of 1,002 that discloses across all nine issues.
The most frequent disclosure was board oversight of AI at 26%. For 119 companies, board oversight is the only AI disclosure.
There is no consensus on who should monitor AI safety
A federal agency similar to the FDA is the single most popular choice among daily AI users (23%), followed by an industry coalition of AI technology leaders (18%) and an international board of industry experts (13%). One in five (20%) say they are not sure.
Daily AI users are notably more likely to favor industry-led oversight – 29% choose an industry coalition (vs. 18% overall) and 18% pick an association of industry leaders overseen by Congress (vs. 11% overall).
Corporate leaders can now explore where they stand and take action to build trust. Through Just Capital’s flagship Just Intelligence product, Russell 1000 companies can now see how their public reporting meet the expectations of the American public, compare to industry peers, and stack up against leading practice.
This study is part of the organization’s ongoing research on how companies can build trust with the American public as they develop and deploy AI.
The survey was conducted online on August 18, 2026 among a national sample of 1,000 US adults age 18 and older, weighted to align with population benchmarks. Because this is an online survey based on a non-probability sample, no estimate of theoretical sampling error can be calculated.
Disclosure data was collected in August 2026 for 1,002 of the largest US-based companies, including nine hyperscalers (Alphabet, Amazon, Anthropic, IBM, Meta, Microsoft, Nvidia, OpenAI, Oracle), across nine AI-related disclosure areas that were selected based on the priorities of the American public: Board oversight of AI, C-suite AI leadership, AI data protection, AI human oversight, AI harm prevention, employee AI upskilling, AI impact on jobs, community AI investments, and anti-surveillance and personal data protection commitment.
Additional details on methodology are available upon request.
