AI Safety has been in the news a lot lately, due in part to rogue AI agents . This has led to industry calls for some form of industry or government imposed standards. In July 2026 Demis Hassabis, Chair of Google DeepMind, wrote on X about the need for an industry-funded, federally overseen AI Standards Body. In September 2026 Dario Amodei, CEO of Anthropic, wrote on his blog about the need for slower-paced AI development in order to continue “building AI at a balanced rate that aims to ensure its safety while still achieving its benefits.” OpenAI has said it supports parts of proposed legislation to establish a federal framework for AI safety. On the other hand, President Trump has expressed clear distaste for AI safety, worrying it will hinder U.S. businesses, a concern shared by others. There is however a business case for AI safety: credible safeguards can support further adoption and follow-on innovation.

Anthropic, Google, OpenAI and other leading AI companies discussing AI safety might sound familiar. In 2023 these same companies and others agreed to a set of voluntary AI safety commitments proposed by the Biden Administration. Some of the 2023 commitments are quite similar to what the companies are now proposing. For example, the first of the 2023 commitments was to “commit to internal and external security testing of their AI systems before their release.” I noted at the time that the voluntary commitments lacked clarity around which external parties would do the testing, and which criteria would be used.

That AI safety is in the news so much lately, and that leading AI companies are now calling for many of the same commitments that were voluntarily agreed upon three years ago, suggest that those 2023 commitments did not have their intended impact. Why should we expect anything different this time around; what, if anything, has changed?

One difference is the increased specificity around how evaluation will be handled. For example, Amodei proposes now that each company “commits to giving ongoing, employee-like access to a team of embedded third-party evaluators.” And Hassabis argues for “an ecosystem of third-party auditors to help with the assessments and development of new benchmarks and evaluations.” We already have some of this infrastructure in place. For example, the U.S. Center for AI Standards and Innovation exists to "facilitate testing and collaborative research related to harnessing and securing the potential of commercial AI systems." METR, a prominent AI safety non-profit organization, describes itself as an independent third-party evaluator of frontier-model risks.

Role of Increased Customer Adoption

One of the biggest differences between 2023 and now is the increase in customer adoption. According to the U.S. Census Bureau, as of late 2023, only about 4% of U.S. businesses had adopted AI. As of September 6, 2026, 23% of U.S. businesses reported using AI, though using revised wording of the AI usage question. Data from other surveys suggest much higher individual level use of AI. The Real-Time Population Survey (RPS), tracked by the St. Louis Fed, reports that, as of August 2026, the share of employed U.S. adults using AI at work is approximately 45 percent (and approximately 62 percent use AI at home or work). As adoption has increased, so too has public interest in AI safety . In this sense, AI is similar to other past technologies, in which the push for coordinated safety standards goes hand-in-hand with rising demand and adoption.

This was the case with electricity , where rapid adoption in the late 1800s and early 1900s led various stakeholders, including insurance, electrical and architectural groups, to establish safety standards. The resulting safety infrastructure helped make broader deployment more trustworthy and scalable, and U.S. household adoption rates grew from single digits in the early 1900s to almost 100% by mid-century. This in turn made it possible for the widespread adoption of other products including refrigerators, washing machines, air conditioners, computers, and other complementary innovations.

History therefore suggests a pro-innovation case for AI safety. Clear standards, credible third-party evaluation, and public accountability can reduce uncertainty for customers and encourage adoption. That, in turn, creates incentives for other companies to develop complementary products and services. There is no guarantee this will happen. But if it does, AI safety will not be a brake on innovation; it will be a foundation for it, allowing the broader AI ecosystem to grow. The challenge for policymakers and industry stakeholders will be to design standards that achieve these safety goals without unnecessarily stifling innovation.