Too much of AI policy today is focusing on things like alignment, 'rogue' AI agents, ‘superintelligence,’ safety testing, transparency reports, pre-clearance for frontier models, preventing 'extinction events' and the ravings, whims, and predictions of billionaire executives. While some of the issues are important and should be dealt with seriously, they engage on the terms of the AI companies and fail to focus on people  and the very real damage AI companies are doing today. Just in the last few weeks, we’ve seen firsthand in the latest news cycle, how the focus on vague concerns and building up the technology as ungovernable leads to Congressional leaders deferring to the companies who have always prioritized profits over people, or focusing on science fiction narratives from companies desperate to secure distraction.

The focus on these policies crowd out AI policy rooted in people's lived experiences – like having your data abused, your loved ones scammed, feeling powerless against deepfakes of you and your family, being constantly watched on every street corner, having AI systems make unreliable and unaccountable decisions about you, having AI constantly change the price of necessities because you logged on at the wrong time or searched the wrong thing, and having chatbots misrepresent some sort of professional expertise when you’re in a time of need. 

AI policy focusing on doomsday outcomes and “alignment” obfuscates the underlying fact that AI systems are products. They are products built by people and companies, used and integrated into systems by other people, and can be designed in ways that don’t threaten people’s privacy and liberty or important technological innovations. Those people and companies must be held accountable for the decisions they make – even when those decisions lead to a story that sounds like autonomous, uncontrollable AI robots are coming for us. 

Kitchen Table AI policy ideas center around privacy, personal autonomy, fairness, safety, affordability, and sanity. The policies below give lawmakers and voters a concrete way to communicate what they stand for and want, at both the state and federal levels. With midterm elections top of mind, most of these policy ideas have polled incredibly well, and all have strong legislative language prepped and ready to go (relevant bills are linked in titles below)

• Pass a strong consumer privacy law. Get rid of 1) complicated forms that result in understandable apathy, 2) problems understanding what data is being collected and how it is being used and 3) companies collecting, keeping and using personal data for no (or for nefarious) reasons. And even when they do have a reason - they should only be able to use it for that reason! No more opt-outs on platforms you used for 15 years because now they want to get into the AI game – this is an unfair practice that must be strictly opt-in with adequate information for consumers.

• Specifically, turn off the training-data firehose and open up the black box. Companies should not be allowed to secretly and nonconsensually use people’s inputs, work, photos, or other data to train AI systems. Providers should have to disclose the types and sources of training data they use, both to protect people and to ensure fair competition.

• Cut off the kerosene for the raging scam wildfire. We need to ban data brokers’ sale of  personal data, and AI companies should have a duty of care not to produce scam or fraud content. Further, social media platforms must be required to stem the flow of scams on their platforms. This should include a clarification that Section 230 does not apply to paid advertisements.

• Prohibit mass surveillance wearable technologies. They are creepy and bad -- and in the age of AI, the tech oligarchs see "always on" glasses, headphones, pens, necklaces and more as the next way to continue to get training  and increasingly sensitive data for advertising targeting, manipulation capabilities, and for law enforcement purposes. Medical devices for people with disabilities that happen to use them shouldn’t be included in this, but, at the same time, they do not need to be infused with FaceID algorithms or hosted by Meta.

• Set clear limits on government use of AI. Public agencies should face rigorous transparency requirements for any AI system they use or contract for. Governments should not buy third-party data to track people or support law enforcement surveillance. And schools should not become testing grounds for AI systems before basic safeguards exist.

• Hold companies and users accountable for nonconsensual intimate imagery. Liability should attach not only to the individual who prompts or uses a tool to generate abusive intimate images (such as in TAKE IT DOWN) but also to the model creator or AI company that facilitates their creation.

• Protect people from nonconsensual recording and synthetic media. There should be clear prohibitions on creating or distributing audio, photos, or video of another person without consent, along with authentication requirements that make it harder to pass off fake or manipulated media as real.

• Preserve the ability of states to pass protective tech laws – reject all forms of moratoria. Blocking states from protecting their residents is undemocratic, unfair, and anti-innovation. States have long served as laboratories of consumer protection, and they should remain able to respond when federal policy falls short.

• Pass “One Fair Price” laws. People should not be subjected to whatever surveillance-based, personalized, dynamic, or collusive pricing practice an industry adopts next. A fair price should not depend on what a company’s data profile says it can extract from you.

• Pause data center development until strict rules are in place. Communities should have a real say in what is built around them. Data center developers should not receive tax breaks or taxpayer subsidies for facilities that strain local resources. Ratepayers should not be stuck with the bill. And companies should be required to report transparently on environmental impacts, energy use, water use, and promised jobs, with clear benchmarks and clawbacks.

• Protect users from exploitative AI “agents” with a fiduciary standard and clear data processing limits. If a company wants to advertise being your assistant by making shopping choices, representing you in decision-making realms, and more, they need to be held to a standard ensuring they act in your best interest. 

• Pass algorithmic decision-making limits, transparency, and accountability requirements. For years, people have had decisions about jobs, loans, housing, and more made by opaque and unaccountable AI systems. As adoption of these systems gets increasingly pushed in government and corporate America, this problem is getting worse. 

• Ban AI toys. Children should not be used as test subjects for always-on, data-hungry products designed to simulate companionship, harvest attention, or normalize surveillance in the home.

• Clarify liability for AI medical and legal impersonation. Chatbot companies should be accountable when their products falsely represent themselves as qualified professionals or provide medical advice without appropriate safeguards.

• Strengthen consumer protection agencies throughout the economy. At the state level, that means building up privacy and consumer protection offices, including local and county-level enforcement capacity. At the federal level, it means revitalizing the FTC and CFPB and restoring their independence. AI policy will only work if the agencies charged with enforcing the rules have the resources, authority, and backbone to do it.

Decisionmakers at every level of government should focus on AI policies that address everyday experiences and concerns, and ensure they remain independent of the AI companies regardless of the election outcome and whom the companies hire to be lobbyists. As always, CFA remains a resource for decisionmakers and other groups looking to put these policies into action.