DX Today AI Daily Brief - Monday, March 23, 2026

Today's top AI stories: White House advisor Kratsios outlines AI framework for Congress urging state law preemption. Wall Street skepticism persists post-Nvidia GTC despite Goldman's bullish $250 target. Microsoft rolls back Copilot AI integrations across Windows 11. Anthropic-Pentagon legal battle deepens as agencies remain in limbo. Hachette pulls horror novel Shy Girl over AI-generated text evidence. Jensen Huang proposes AI tokens as new engineer compensation. Meta plans 20% layoffs to offset up to $135B in AI spending. Cerebras and AWS partner for fastest cloud AI inference. Morgan Stanley warns AI breakthrough is imminent with looming power crisis. US military confirms advanced AI tools in Iran operations. OpenAI launches GPT-5.4 mini and nano models. State AI legislation advances across Georgia, Washington, and North Carolina.

Transcript

SPEAKER_02 0:00

It's Monday, March 23, 2026. You're listening to the DX Today AI Daily Brief. This morning, the White House lays out its vision for AI regulation. Wall Street pushes back on NVIDIA's grand ambitions. Microsoft dials back its co-pilot push, and a major publisher pulls a novel over AI suspicions. Let's get into it.

SPEAKER_04 0:26

White House science advisor Michael Kratzios appeared on Fox News Sunday to discuss the administration's AI policy framework sent to Congress late last week. The plan urges lawmakers to preempt state AI regulations, arguing that a patchwork of local rules would stifle American innovation. The framework calls for no new federal agencies to oversee AI, instead, relying on existing regulators for sector-specific guidance. It does carve out exceptions, allowing states to maintain laws protecting children, including bans on AI-generated child exploitation material. The proposal has drawn both praise from industry leaders and sharp criticism from state attorneys general who see it as an overreach of federal authority.

SPEAKER_00 1:40

But Wall Street wasn't buying it. The stock dropped during the presentation as investors weighed fears of an AI bubble. Goldman Sachs held firm with its$250 price target and buy rating, while Rosenblatt went even higher at$325. The disconnect between Silicon Valley optimism and market caution reflects what analysts call a great new uncertainty. Investors want revenue proof, not roadmaps.

SPEAKER_02 2:12

Meanwhile, Microsoft is rethinking its approach. Microsoft announced a strategic rollback of its co-pilot AI integrations across Windows 11. The company is removing copilot entry points from photos, widgets, notepad, and the snipping tool. Executive Vice President Pavan Davoluri said the company is becoming more intentional about how and where copilot integrates, focusing on experiences that are genuinely useful rather than ubiquitous. The move follows months of user complaints about what critics have called AI bloat. Microsoft had already shelved deeper system-level copilot features in settings and file explorer. The pivot signals a broader industry recognition that aggressive AI deployment can erode user trust faster than it builds utility. Tensions continue at the Pentagon.

SPEAKER_05 3:17

The Pentagon classified Anthropic as a supply chain risk after the company insisted its AI models not be used for autonomous weapons or domestic surveillance. That designation effectively bars defense contractors from using Claude in any government work. Anthropic filed suit, calling the government's actions unprecedented and unlawful. Meanwhile, a Pentagon memo has opened the door for exemptions beyond the original six-month phase-out period, suggesting some defense leaders are reluctant to lose access to models they had previously praised as superior. Federal agencies remain in limbo as the dispute plays out in court.

SPEAKER_03 4:23

Content analysis by AI detection firm Pangram indicated that 78% of the book was likely AI generated, flagging repetitive phrases and formatting anomalies. Ballard maintains she did not use AI herself, claiming an editor she hired used AI tools on an early draft without her knowledge. The New York Times called it the first commercial novel from a major publisher to be retracted over AI involvement. The case raises urgent questions about verification in an industry built on original human expression.

SPEAKER_02 5:00

A new form of compensation takes shape.

SPEAKER_04 5:10

At GTC, Jensen Huang proposed giving engineers roughly half their base salary in computational tokens, the units that power tools like Claude, ChatGPT, and Gemini. By his math, top engineers could burn through$250,000 a year in AI compute. Venture capitalist Tomas Tungus of Theory Ventures framed it as a fourth pillar of compensation, alongside salary, bonus, and equity. But engineers are cautious. A large token allotment comes with expectations of dramatically higher output. As one analyst noted, if you are burning 100,000 in tokens, you had better be eight times more productive.

SPEAKER_02 5:53

Big tech faces a balancing act.

SPEAKER_00 5:56

Meta is reportedly planning to cut up to 20% of its workforce as it struggles to offset surging AI costs. The company projected capital expenditure between$115 and$135 billion for 2026, roughly double what it spent last year. CEO Mark Zuckerberg has called this a pivotal year for AI, aligning with his vision of building personal superintelligence. The layoffs would be met as most significant since its 2022 restructuring. Investors responded positively, pushing the stock up 3% on the news, reflecting a market that rewards cost discipline even as it questions the return on massive AI investments.

SPEAKER_01 6:51

Cerebrus Systems and Amazon Web Services announced a multi-year collaboration to bring the fastest AI inference to the cloud. AWS will deploy Cerebrus CS3 systems in its data centers, accessible through Amazon Bedrock. The partnership introduces a disaggregated architecture, pairing AWS trainium chips for prefill with Cerebrus waferscale engines for decode, delivering what both companies claim is five times more high-speed token capacity. Cerebrus already powers inference for OpenAI, Cognition, and Meta at speeds reaching 3,000 tokens per second. AWS Vice President David Brown said the collaboration targets the critical bottleneck of real-time inference for coding assistance and interactive applications.

SPEAKER_05 7:44

Morgan Stanley is warning that a massive AI breakthrough is imminent, and most of the world is not ready for it. The bank's latest report projects nearly$3 trillion in global AI infrastructure investment flowing through the economy by 2028, with more than 80% of that spending still ahead. The bank forecasts a net U.S. power shortfall of 9 to 18 gigawatts through 2028, a 12 to 25% deficit. On the labor front, executives are already executing large-scale workforce reductions driven by AI efficiencies. OpenAI's GPT 5.4 thinking model scored 83% on the GDP Val benchmark, at or above human expert level on economically valuable tasks.

SPEAKER_02 8:34

AI on the battlefield draws scrutiny.

SPEAKER_03 8:38

The U.S. military has confirmed using advanced AI tools in its operations against Iran, now in their fourth week. Central Command Leader Brad Cooper said AI systems enable troops to analyze extensive data sets in seconds, helping commanders filter information and make faster decisions. The military is using AI across the targeting cycle, from strategy meetings to battlefield execution, including logistics, drone navigation, and data fusion from multiple sensors. Human rights groups have raised concerns about civilian casualties, with Iranian officials reporting over 1,200 fatalities. Cooper emphasized that humans retain final authority over the use of force, but critics note the boundaries of AI's role in combat remain dangerously unclear.

SPEAKER_02 9:31

OpenAI expands its model lineup.

SPEAKER_04 9:34

OpenAI has launched GPT 5.4 Mini and GPT 5.4 nano, smaller and faster versions of its flagship model, optimized for coding, tool use, and high-volume workloads. The Mini model is more than two times faster than GPT-5 Mini and approaches full GPT 5.4 performance on software engineering benchmarks. It ships with a 400,000 token context window. The nanovariant is the cheapest model in OpenAI's lineup, priced at 20 cents per million input tokens, targeting classification, data extraction, and lightweight agent tasks. Both models are available in the API, Codex, and ChatGPT. The release reflects a broader industry trend toward efficient, specialized models rather than ever larger monoliths.

SPEAKER_02 10:32

And finally, state legislatures are moving.

SPEAKER_00 10:36

A wave of AI legislation is advancing across state capitals. In Georgia, three bills are progressing before the April 6th adjournment deadline, including a chatbot disclosure and child safety measure, and a bill prohibiting insurance decisions based solely on AI. Washington state is considering at least five AI-related bills covering health insurance decisions, deep fake protections, and content accountability. North Carolina has introduced measures on AI-generated content regulation and companion chatbot oversight. The legislative push comes as the White House framework urges Congress to preempt state AI laws, setting up a potential clash between federal and state authority over who regulates artificial intelligence.