AI Reshapes Regulation, Security, and Science: Key Innovations in 2024

Artificial intelligence is rapidly reshaping industries, from government regulation and financial security to scientific research. This press review explores how a college student is leading AI-driven deregulation at HUD, Microsoft’s compact Phi language models are setting new benchmarks in reasoning, Persona is tackling the internet’s bot crisis with advanced identity verification, and TranscriptFormer is unlocking the secrets of cell biology. Each story highlights the transformative power of AI and its growing influence on policy, business, and science.

AI-Driven Deregulation at HUD: A College Student at the Helm

A significant shift is underway at the U.S. Department of Housing and Urban Development (HUD), where Christopher Sweet, a third-year undergraduate from the University of Chicago, has been appointed as a special assistant within Elon Musk’s Department of Government Efficiency (DOGE). According to WIRED, Sweet’s primary responsibility is to lead an initiative that uses artificial intelligence to review and potentially rewrite HUD’s regulations, with a focus on identifying areas for deregulation.

Sweet, who has a background in economics and data science and is fluent in Portuguese, has been given access to HUD’s Public and Indian Housing Information Center and its enterprise income verification systems. His work involves producing an Excel spreadsheet with approximately a thousand rows, each highlighting policy areas where the AI tool has flagged possible regulatory overreach and suggesting alternative language. Staff from the Office of Public and Indian Housing (PIH) are required to review these AI-generated recommendations and justify any objections.

The AI model, reportedly being refined for broader government use, crawls through the Code of Federal Regulations (eCFR) to identify noncompliant or overly complex rules. The process includes quantifying how many words can be eliminated from individual regulations and providing a percentage of noncompliance, though the calculation method remains unclear. Once reviewed by PIH, recommendations are sent to the Office of the General Counsel for approval.

Sweet’s appointment and the broader deregulation effort align with the Project 2025 policy document, which advocates for significant deregulation across various sectors, including environment and food and drug enforcement. Despite his limited government experience, Sweet is reportedly the lead on the AI deregulation project for the entire administration.

“It all sounds crazy—having AI recommend revisions to regulations,” one HUD source told WIRED. “But I appreciated how much they’re using real people to confirm and make changes.”

Sweet’s professional background includes founding East Edge Securities and working with private equity firms such as Pertento Partners and Tenzing Global Investors. He is also listed as a board member of Paragon Global Investments and is set to join Nexus Point Capital as a private equity summer analyst.

• Christopher Sweet is leading AI-driven regulatory review at HUD.
• AI-generated recommendations are being reviewed by human staff before legal approval.
• The initiative is part of a broader push for government deregulation as outlined in Project 2025.

Source: WIRED

Summary: HUD is leveraging AI, under the leadership of a college student, to streamline and potentially deregulate federal housing policies, with human oversight and legal review as safeguards.

Microsoft’s Phi Small Language Models: A Year of Breakthroughs

Microsoft has marked a year since the introduction of its Phi small language models (SLMs), with the latest releases—Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning—setting new standards for efficient AI. According to Microsoft Azure, these models are designed to deliver advanced reasoning capabilities typically found only in much larger models, but with a fraction of the computational requirements.

Phi-4-reasoning is a 14-billion parameter open-weight model, trained via supervised fine-tuning on curated reasoning demonstrations. It generates detailed reasoning chains and leverages additional inference-time compute. Phi-4-reasoning-plus, an enhanced version, uses 1.5 times more tokens than Phi-4-reasoning and achieves even higher accuracy through reinforcement learning.

Both models outperform OpenAI o1-mini and DeepSeek-R1-Distill-Llama-70B on most benchmarks, including mathematical reasoning and Ph.D.-level science questions. Notably, they surpass the full DeepSeek-R1 model (671 billion parameters) on the AIME 2025 test, the qualifier for the USA Math Olympiad. Phi-4-mini-reasoning, optimized for environments with limited computing power, is trained on over one million diverse math problems and is ideal for educational and edge applications.

Phi models are now integral to Copilot+ PCs, with the NPU-optimized Phi Silica variant preloaded for fast, power-efficient performance. These models are used in applications like Outlook’s Copilot summary features, even offline, and are available as developer APIs for integration into various productivity tools.

Microsoft emphasizes responsible AI development, employing supervised fine-tuning, direct preference optimization, and reinforcement learning from human feedback to ensure safety, fairness, and reliability. The company acknowledges that all AI models have limitations and provides detailed model cards outlining responsible AI practices.

• Phi-4-reasoning and Phi-4-reasoning-plus outperform much larger models on key benchmarks.
• Phi-4-mini-reasoning is optimized for low-latency, resource-constrained environments.
• Phi models are integrated into Windows 11 devices and Copilot+ PCs.

Source: Microsoft Azure

Summary: Microsoft’s Phi small language models deliver high-level reasoning in compact, efficient packages, setting new standards for AI performance and integration across devices.

Persona: Battling the Internet’s Bot Crisis with AI-Powered Identity Verification

The proliferation of bots on the internet has reached a critical point, with bots accounting for about 50% of all internet traffic—a figure projected to rise to 90% by 2030 as AI agents automate more online tasks. Forbes reports that this surge presents major challenges for companies seeking to distinguish between legitimate users and malicious actors, especially as AI makes it easier to mimic voices, faces, and create fake IDs.

San Francisco-based Persona, now valued at $2 billion after raising $200 million in Series D funding, is at the forefront of this battle. The company provides identity verification solutions to 3,000 clients, including OpenAI, LinkedIn, Etsy, Reddit, DoorDash, and Robinhood. Persona’s technology combines multiple verification methods—such as government ID uploads, selfies, video recordings, and NFC passport scans—with machine learning models that analyze user behavior and risk signals.

Persona’s approach is highly customizable, creating verification “flows” tailored to each client’s needs and risk profiles. For example, OpenAI uses Persona to screen hundreds of millions of users across 225 countries, ensuring compliance with international watchlists and sanctions. Coursera verifies its 168 million users in 200 countries, adjusting the process based on the type of course accessed.

The bot problem is not just about malicious actors; good bots accounted for 14% of automated traffic in 2024, while malicious bots were responsible for 37%. U.S. companies lose between $18 billion and $31 billion annually to AI-based attacks, and global losses from bot attacks range from $68 billion to $116 billion, according to Imperva.

Persona’s data is stored on AWS and GCP servers, managed by client companies. The company has faced legal scrutiny over data privacy, with lawsuits in 2024 alleging unauthorized collection and use of biometric data, though these were dismissed or denied by Persona.

• Bots make up 50% of internet traffic, projected to reach 90% by 2030.
• Persona’s identity verification is used by major tech companies worldwide.
• Annual losses from AI-based and bot attacks are in the tens of billions of dollars.

Source: Forbes

Summary: Persona is addressing the escalating bot crisis with advanced, customizable AI-powered identity verification, helping major companies secure their platforms amid rising online threats.

TranscriptFormer: AI Advancements in Cell Biology

The intersection of artificial intelligence and cell biology is exemplified by TranscriptFormer, a new AI model developed to help scientists explore cross-species biology at single-cell resolution. According to the Chan Zuckerberg Initiative, TranscriptFormer is designed to decode the unique “sentence” of gene expression in each cell, providing insights into how cells differentiate and function in both healthy and disease states.

The development team, comprising AI researchers, engineers, biologists, and data scientists, collaborated closely to train the model on over a hundred million cells. The goal is to understand the differences between cells, despite each having the same DNA, by analyzing which genes are expressed in various cell types—such as brain, liver, or muscle cells.

“If you imagine each gene like a word, each cell has its own sentence of words that make it unique. We are building technology to decode these sentences—for every cell, in healthy and disease states, across organisms—so we can understand this very fundamental part of biology.” — James Pearce, AI researcher

TranscriptFormer represents a significant step forward in leveraging AI to unlock the complexities of biology, with the potential to accelerate the discovery of new treatments for human diseases.

• TranscriptFormer is trained on over 100 million cells.
• The model decodes gene expression patterns at single-cell resolution.
• Collaboration between AI experts and biologists is central to its development.

Source: Chan Zuckerberg Initiative

Summary: TranscriptFormer is a pioneering AI model that enables detailed analysis of gene expression in individual cells, advancing the understanding of cell differentiation and disease mechanisms.

Einschätzung der Redaktion

The appointment of a college student to lead an AI-driven regulatory overhaul at a major federal agency is a striking signal of both the rapid pace of technological adoption and a willingness to disrupt traditional bureaucratic hierarchies. This move demonstrates a bold confidence in the capabilities of artificial intelligence to streamline complex regulatory frameworks, but it also raises critical questions about oversight, accountability, and the depth of institutional knowledge required for such high-stakes policy work. The integration of AI into regulatory review processes could set a precedent for other government agencies, potentially accelerating deregulatory efforts and reshaping the balance between efficiency and thoroughness in public administration. However, the reliance on a relatively inexperienced individual at the helm underscores the experimental nature of this initiative and highlights the need for robust safeguards to ensure that technological innovation does not outpace prudent governance.

• AI is being positioned as a transformative tool in federal regulatory processes.
• The leadership choice reflects a disruptive approach to traditional government roles.
• Potential for significant impact on future regulatory and deregulatory strategies.

Sources:

• DOGE Put a College Student in Charge of Using AI to Rewrite Regulations
• One year of Phi: Small language models making big leaps in AI
• AI Is Making The Internet’s Bot Problem Worse. This $2 Billion Startup Is On The Front Lines
• Meet Scientists Behind TranscriptFormer, a New AI Model
• Control the Composition of AI-Generated Images With the NVIDIA AI Blueprint for 3D-Guided Generative AI
• The Strange Physics That Gave Birth to AI