A New York Times investigation published this week titled “Silicon Valley Is Bracing for a Permanent Underclass” documents something the AI industry has kept carefully managed: the people building these systems hold far bleaker views about what they will do to workers than their public statements suggest.
Sources inside frontier AI labs told the reporter, Jasmine Sun, they expressed “more extreme concern about the labor market impacts of AI in private conversation, but suddenly became optimists once I turned on the mic.”
Also on Apple Podcasts
The piece names this the San Francisco consensus.” The Consensus is a shared private belief among engineers, venture capitalists, and founders that advanced AI will displace millions of jobs, concentrate wealth in AI companies and capital owners, and produce lasting inequality.
Anthropic CEO Dario Amodei has said publicly that AI may create an unemployed or very-low-wage underclass. Sam Altman predicted in 2021 that without aggressive policy action, most people will end up worse off than they are today. These are the optimists. They are the ones willing to speak for attribution.
The investigation also documents the institutional mechanism that keeps the public conversation sunnier than the private one. When a veteran lobbyist joined OpenAI’s leadership in April 2024, the company deprioritized research that could produce unflattering findings. OpenAI has excluded public discuss of studies on the gender gap in AI’s labor market effects and on long-run economic forecasting, among others.
The company’s messaging shifted toward GDP growth stories. The progressive proposals in OpenAI’s April white paper are real ideas. Nevertheless, the absence of any specific legislation the company has committed to speaks volumes.
Get the eBook that exposes the Narratives Tech uses to build its AI Empire. $4.95 SPECIAL SALE PRICE for PDF or ePUB formats; no subscription required. 3.5-hr reading time.
The people building artificial intelligence did not invent their ideas. They inherited them.
The people with the clearest understanding of the problem are the people whose economic position makes pessimism about it most costly. For instance, Anthropic’s annualized revenue has surged to $30 billion, driven by Claude Code, an agent that automates knowledge work. The rewards for optimism, in other words, are tremendous; transparence, not so much.
Palantir CEO Alex Karp named the underlying concern plainly to a panel audience in March. “The biggest challenge to AI in this country is political unrest. If I were sitting here in private with my peers, I’d be telling them the country could blow up politically and none of us are going to make any money when the country blows up.”
Karp frames AI impact as a risk management issue rather than worker protection problem. That’s is precisely what makes the Connecticut legislation passed last Friday so significant.
Connecticut’s Senate Bill 5, now the Connecticut Artificial Intelligence Responsibility and Transparency Act, cleared the House 131-17 and the Senate 32-4. Governor Lamont confirmed he will sign it, reversing his position from 2025.
The bill requires employers to disclose when AI tools are used in hiring, promotion, discipline, or termination. It amends anti-discrimination law to confirm that deploying an AI tool is not a legal defense against a discrimination claim.
Get the eBook that examines how the AI invasion already happened. You just weren’t invited. $9.95 flat fee for PDF or ePUB; no subscription required. 2-hr reading time.
The provision with the broadest labor market implications takes effect October 1, 2026. Any employer filing a federal WARN notice in Connecticut must disclose whether the layoffs are related to AI or technological change. The Worker Adjustment and Retraining Notification Act of 1988 is a U.S. labor law that protects employees, their families, and communities by requiring most employers with 100 or more employees to provide notification 60 calendar days in advance of planned closings and mass layoffs of employees.
That single requirement could begin to close the measurement gap the AI Labor Report has documented throughout 2026. Connecticut passed its bill by a margin of 131-17.
Meanwhile, a federal executive order directs the attorney general to challenge state AI legislation as an obstacle to national policy.
The contrast between the State and Federal approaches is sharpest when set against what Italy has already put in place.
Italy’s Law No. 132/2025 became fully operational in October 2025. The law bans fully automated employment decisions. Every employment-related decision made with AI input requires meaningful human oversight.
Before deploying any AI system in the workplace, employers must consult with trade unions. Workers receive advance disclosure when AI tools are used in hiring and evaluation and retain the right to challenge AI-driven decisions and request human review.
Italy is the first EU member state with comprehensive national AI legislation. Its union consultation requirement is the strongest worker protection provision currently in force anywhere in the developed world.
Future Forwarded is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
The gap between the Italian framework and the American federal posture is a gap in whose interests the regulatory architecture is designed to serve. Italy requires employers to consult trade unions before deploying AI that affects workers. The United States federal government is working to prevent states from requiring employers to disclose whether AI caused a layoff.
The San Francisco consensus says the disruption is coming. The policy question is whether the institutions making decisions about it are building the regulatory infrastructure that protects the people most exposed to the disruption — or the systems that protect the people who are accountable for the AI rollout.
Connecticut and Italy chose one answer; The federal government is choosing another.
Get full access to Future Forwarded at futureforwarded.substack.com/subscribe
