Large U.S. companies just hit the brakes on AI—adoption fell from 14% to 12% in two months, the first decline since tracking began. MIT research explains why: 95% of enterprise pilots deliver zero ROI. The gap between AI hype and workflow reality is widening.
Good Morning from San Francisco, Oracle just inked a $300 billion computing deal with OpenAI. That's not a
Oracle bets $300B on OpenAI's computing future, but the math is stark: OpenAI generates $10B annually while committing to $60B yearly. The deal either transforms Oracle into an AI infrastructure leader—or becomes a cautionary dot-com tale.
The Cruz proposal offers a laboratory for testing whether innovation imperatives can coexist with democratic oversight. Whether Congress embraces this model may determine how America balances technological competition with institutional governance in the AI era.
💡 TL;DR - The 30 Seconds Version
📜 Cruz introduced the SANDBOX Act allowing AI companies to seek 2-year regulatory waivers renewable up to 10 years from federal agencies.
⏰ Waivers get automatically approved if agencies don't respond within 90 days, with OSTP having override power to reverse denials.
🏛️ The bill excludes state regulation preemption—the provision that killed Cruz's 10-year moratorium attempt in July (defeated 99-1).
⚠️ Consumer groups call it "experimenting on Americans" while tech industry group NetChoice endorsed the "innovation-first approach."
🇨🇳 Cruz frames the policy as essential for beating China in AI development, echoing Trump's technological competition priorities.
🔬 Regulatory sandboxes show mixed results globally—the UK's financial sandbox enabled innovation but also created exploitable gaps.
Tech gets temporary regulatory relief, consumer groups cry foul. Federal override mechanism sparks separation of powers questions
Republican Senator Ted Cruz introduced legislation Wednesday that would allow AI companies to seek temporary waivers from federal regulations, marking the first concrete policy proposal from his broader framework to ensure American dominance in artificial intelligence development.
The SANDBOX Act would permit companies to apply for two-year exemptions from "obstructive regulations" affecting AI products, renewable up to a decade. Companies must outline safety and financial risks plus mitigation plans. If agencies don't respond within 90 days, waivers get automatically approved—a provision that's drawing sharp criticism from consumer advocates who call it "experimenting on Americans."
The proposal emerges from Cruz's Commerce Committee chairmanship and dovetails with President Trump's AI Action Plan, which called for regulatory sandboxes among its 90-plus recommendations. It also represents a tactical retreat from Cruz's failed attempt to impose a 10-year moratorium on state AI regulation, which the Senate defeated 99-1 in July.
The bill's most controversial element grants the White House Office of Science and Technology Policy unprecedented power to reverse agency denials of waiver requests. OSTP Director Michael Kratsios endorsed the approach during Wednesday's hearing, citing successful drone sandbox programs that helped the FAA develop new regulations.
But the override mechanism creates potential separation of powers tensions. Federal agencies possess subject-matter expertise in their domains—the FTC in consumer protection, FDA in health safety, Treasury in financial stability. The bill would allow OSTP, a White House political office, to overrule career regulators' safety determinations.
Consumer advocacy group Public Citizen argues this concentrates too much power in political appointees. "Companies that build untested, unsafe AI tools could get hall passes from the very rules designed to protect the public," said J.B. Branch, the group's Big Tech accountability advocate.
From industry's perspective, the current regulatory landscape reflects what Cruz calls "ossified" rules. A 2017 Deloitte survey found 68 percent of federal regulations never get updated, potentially blocking innovations that didn't exist when the rules were written. Cruz's bill tries splitting the difference—keep regulations intact while letting companies test around them.
Cruz's bill notably omits any preemption of state AI regulations—the provision that sank his earlier moratorium effort. States from California to New York are advancing their own AI oversight frameworks, creating what industry calls a compliance nightmare.
Delaware already operates its own AI sandbox program, which Senator Lisa Blunt Rochester highlighted during Wednesday's hearing. But she questioned whether OSTP represents "the appropriate place" for a federal sandbox program. Even Democrats who like sandboxes don't want them run from the White House.
California's AI safety bill keeps moving despite tech lobbying. New York just passed its own "catastrophic harm" legislation. States are filling a vacuum that Congress hasn't touched.
Kratsios told Cruz that "a patchwork of state regulations is anti-innovation," particularly harmful to smaller firms lacking compliance resources. He urged Congress to "revisit the possibility of preempting state laws." The dynamic reveals ongoing tension between innovation imperative and federalism principles.
Cruz positions his framework explicitly as a response to Chinese AI ambitions. "If China wins the AI race, the world risks an order built on surveillance and coercion," he argued when releasing his five-pillar AI policy framework alongside the SANDBOX Act.
This framing echoes the successful internet policy model that Cruz credits with American digital dominance over Europe. While the EU pursued "heavy-handed prior approval regulatory approaches," America adopted lighter-touch policies that enabled tech and energy-led innovation.
The China argument cuts multiple ways, though. Loosening safety rules could enable exactly what authoritarian regimes do—mass surveillance, social control, algorithmic manipulation. America wants to win the race while keeping its values. That's not simple.
The broader geopolitical context shows Cruz aligning with Trump's technological competition priorities. Whether Democrats buy this framing remains unclear, especially given their focus on AI safety and transparency.
Cruz's bill requires yearly reports to Congress on which waivers got granted and how they worked out. Companies would still face existing laws on fraud, discrimination, and consumer protection—this covers regulatory hurdles, not criminal behavior.
But regulatory sandboxes have mixed track records. The UK's financial services sandbox gets cited as a success story, yet it's also created gaps that sophisticated players exploit. Sometimes waivers help regulators learn. Sometimes they just delay necessary oversight.
Tech industry support appears strong. NetChoice, representing Amazon, Google, and Meta, endorsed Cruz's "American innovation-first approach." But notable silence from leading AI companies suggests they're waiting to see whether the bill gains bipartisan momentum before fully embracing it.
Why this matters:
• Federal AI policy increasingly reflects geopolitical competition frames, potentially prioritizing speed over safety considerations
• The OSTP override mechanism creates precedent for political appointees reversing technical agency determinations across regulatory domains
Q: What specific regulations could AI companies get waivers from?
A: Any federal rule affecting AI products—health data privacy (HIPAA), financial market stability (SEC rules), transportation safety (DOT regulations), or consumer protection (FTC guidelines). Companies must prove the regulation blocks their AI work and outline risk mitigation plans.
Q: What happens if a company causes harm during their waiver period?
A: Companies still face existing criminal laws on fraud, discrimination, and consumer protection. The sandbox only waives regulatory compliance, not legal liability. However, the bill doesn't specify penalties for companies that violate their risk mitigation agreements.
Q: How does this compare to existing regulatory sandboxes globally?
A: Over 50 countries run sandboxes, starting with the UK's 2016 financial services program. Wyoming and Arizona pioneered US sandboxes in 2019, followed by Utah's all-industry program in 2021. Most focus on specific sectors, unlike Cruz's cross-agency approach.
Q: Which federal agencies would be most affected by waiver requests?
A: Likely the FTC (consumer protection), FDA (healthcare AI), SEC (financial algorithms), DOT (autonomous vehicles), and Treasury (fintech). Each agency decides on waivers in their domain, but OSTP can override denials—a power no existing sandbox program grants.
Q: What are the chances this bill actually becomes law?
A: Uncertain. Republicans control the Senate, and the White House supports it. But Democrats emphasize AI safety over speed, and Cruz's 10-year state regulation moratorium failed 99-1 in July. The bill's survival likely depends on Democratic willingness to compromise on oversight mechanisms.
Marcus Schuler
Marcus Schuler
China's surveillance tech, originally built with IBM and other US companies in the 2000s, is now being sold as 'control-tech-as-a-service' globally. Geedge Networks ships turnkey censorship systems to Pakistan, Myanmar, and others—all while avoiding sanctions.
Nvidia brands Congressional chip hawks as "AI doomers" while supply constraints limit growth regardless of policy outcomes. Chinese firms still want chips despite Beijing pressure. Physical bottlenecks reshape the AI hardware landscape.
California's SB 53 advances with Anthropic's surprise endorsement, splitting the AI industry on transparency rules. The bill requires safety disclosures and incident reporting for frontier models, creating a state template as Congress remains stalled.
Trump's new chip diplomacy: annual permits replace permanent waivers for Samsung and SK Hynix in China. Washington gains oversight, Seoul gets continuity with red tape. But the real bottleneck isn't logic—it's memory production.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
